diff --git a/.gitignore b/.gitignore new file mode 100644 index 0000000..ac26bbc --- /dev/null +++ b/.gitignore @@ -0,0 +1,163 @@ +*/.vscode/* +# Byte-compiled / optimized / DLL files +__pycache__/ +*.py[cod] +*$py.class + +# C extensions +*.so + +# Distribution / packaging +.Python +build/ +develop-eggs/ +dist/ +downloads/ +eggs/ +.eggs/ +lib/ +lib64/ +parts/ +sdist/ +var/ +wheels/ +share/python-wheels/ +*.egg-info/ +.installed.cfg +*.egg +MANIFEST + +# PyInstaller +# Usually these files are written by a python script from a template +# before PyInstaller builds the exe, so as to inject date/other infos into it. +*.manifest +*.spec + +# Installer logs +pip-log.txt +pip-delete-this-directory.txt + +# Unit test / coverage reports +htmlcov/ +.tox/ +.nox/ +.coverage +.coverage.* +.cache +nosetests.xml +coverage.xml +*.cover +*.py,cover +.hypothesis/ +.pytest_cache/ +cover/ + +# Translations +*.mo +*.pot + +# Django stuff: +*.log +local_settings.py +db.sqlite3 +db.sqlite3-journal + +# Flask stuff: +instance/ +.webassets-cache + +# Scrapy stuff: +.scrapy + +# Sphinx documentation +docs/_build/ + +# PyBuilder +.pybuilder/ +target/ + +# Jupyter Notebook +.ipynb_checkpoints + +# IPython +profile_default/ +ipython_config.py + +# pyenv +# For a library or package, you might want to ignore these files since the code is +# intended to run in multiple environments; otherwise, check them in: +# .python-version + +# pipenv +# According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control. +# However, in case of collaboration, if having platform-specific dependencies or dependencies +# having no cross-platform support, pipenv may install dependencies that don't work, or not +# install all needed dependencies. +#Pipfile.lock + +# poetry +# Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control. +# This is especially recommended for binary packages to ensure reproducibility, and is more +# commonly ignored for libraries. +# https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control +#poetry.lock + +# pdm +# Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control. +#pdm.lock +# pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it +# in version control. +# https://pdm.fming.dev/latest/usage/project/#working-with-version-control +.pdm.toml +.pdm-python +.pdm-build/ + +# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm +__pypackages__/ + +# Celery stuff +celerybeat-schedule +celerybeat.pid + +# SageMath parsed files +*.sage.py + +# Environments +.env +.venv +env/ +venv/ +ENV/ +env.bak/ +venv.bak/ + +# Spyder project settings +.spyderproject +.spyproject + +# Rope project settings +.ropeproject + +# mkdocs documentation +/site + +# mypy +.mypy_cache/ +.dmypy.json +dmypy.json + +# Pyre type checker +.pyre/ + +# pytype static type analyzer +.pytype/ + +# Cython debug symbols +cython_debug/ + +# PyCharm +# JetBrains specific template is maintained in a separate JetBrains.gitignore that can +# be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore +# and can be added to the global gitignore or merged into this file. For a more nuclear +# option (not recommended) you can uncomment the following to ignore the entire idea folder. +#.idea/ diff --git a/LICENSE.txt b/LICENSE.txt new file mode 100644 index 0000000..f288702 --- /dev/null +++ b/LICENSE.txt @@ -0,0 +1,674 @@ + GNU GENERAL PUBLIC LICENSE + Version 3, 29 June 2007 + + Copyright (C) 2007 Free Software Foundation, Inc. + Everyone is permitted to copy and distribute verbatim copies + of this license document, but changing it is not allowed. + + Preamble + + The GNU General Public License is a free, copyleft license for +software and other kinds of works. + + The licenses for most software and other practical works are designed +to take away your freedom to share and change the works. By contrast, +the GNU General Public License is intended to guarantee your freedom to +share and change all versions of a program--to make sure it remains free +software for all its users. We, the Free Software Foundation, use the +GNU General Public License for most of our software; it applies also to +any other work released this way by its authors. You can apply it to +your programs, too. + + When we speak of free software, we are referring to freedom, not +price. Our General Public Licenses are designed to make sure that you +have the freedom to distribute copies of free software (and charge for +them if you wish), that you receive source code or can get it if you +want it, that you can change the software or use pieces of it in new +free programs, and that you know you can do these things. + + To protect your rights, we need to prevent others from denying you +these rights or asking you to surrender the rights. Therefore, you have +certain responsibilities if you distribute copies of the software, or if +you modify it: responsibilities to respect the freedom of others. + + For example, if you distribute copies of such a program, whether +gratis or for a fee, you must pass on to the recipients the same +freedoms that you received. You must make sure that they, too, receive +or can get the source code. And you must show them these terms so they +know their rights. + + Developers that use the GNU GPL protect your rights with two steps: +(1) assert copyright on the software, and (2) offer you this License +giving you legal permission to copy, distribute and/or modify it. + + For the developers' and authors' protection, the GPL clearly explains +that there is no warranty for this free software. For both users' and +authors' sake, the GPL requires that modified versions be marked as +changed, so that their problems will not be attributed erroneously to +authors of previous versions. + + Some devices are designed to deny users access to install or run +modified versions of the software inside them, although the manufacturer +can do so. This is fundamentally incompatible with the aim of +protecting users' freedom to change the software. The systematic +pattern of such abuse occurs in the area of products for individuals to +use, which is precisely where it is most unacceptable. Therefore, we +have designed this version of the GPL to prohibit the practice for those +products. If such problems arise substantially in other domains, we +stand ready to extend this provision to those domains in future versions +of the GPL, as needed to protect the freedom of users. + + Finally, every program is threatened constantly by software patents. +States should not allow patents to restrict development and use of +software on general-purpose computers, but in those that do, we wish to +avoid the special danger that patents applied to a free program could +make it effectively proprietary. To prevent this, the GPL assures that +patents cannot be used to render the program non-free. + + The precise terms and conditions for copying, distribution and +modification follow. + + TERMS AND CONDITIONS + + 0. Definitions. + + "This License" refers to version 3 of the GNU General Public License. + + "Copyright" also means copyright-like laws that apply to other kinds of +works, such as semiconductor masks. + + "The Program" refers to any copyrightable work licensed under this +License. Each licensee is addressed as "you". "Licensees" and +"recipients" may be individuals or organizations. + + To "modify" a work means to copy from or adapt all or part of the work +in a fashion requiring copyright permission, other than the making of an +exact copy. The resulting work is called a "modified version" of the +earlier work or a work "based on" the earlier work. + + A "covered work" means either the unmodified Program or a work based +on the Program. + + To "propagate" a work means to do anything with it that, without +permission, would make you directly or secondarily liable for +infringement under applicable copyright law, except executing it on a +computer or modifying a private copy. Propagation includes copying, +distribution (with or without modification), making available to the +public, and in some countries other activities as well. + + To "convey" a work means any kind of propagation that enables other +parties to make or receive copies. Mere interaction with a user through +a computer network, with no transfer of a copy, is not conveying. + + An interactive user interface displays "Appropriate Legal Notices" +to the extent that it includes a convenient and prominently visible +feature that (1) displays an appropriate copyright notice, and (2) +tells the user that there is no warranty for the work (except to the +extent that warranties are provided), that licensees may convey the +work under this License, and how to view a copy of this License. If +the interface presents a list of user commands or options, such as a +menu, a prominent item in the list meets this criterion. + + 1. Source Code. + + The "source code" for a work means the preferred form of the work +for making modifications to it. "Object code" means any non-source +form of a work. + + A "Standard Interface" means an interface that either is an official +standard defined by a recognized standards body, or, in the case of +interfaces specified for a particular programming language, one that +is widely used among developers working in that language. + + The "System Libraries" of an executable work include anything, other +than the work as a whole, that (a) is included in the normal form of +packaging a Major Component, but which is not part of that Major +Component, and (b) serves only to enable use of the work with that +Major Component, or to implement a Standard Interface for which an +implementation is available to the public in source code form. A +"Major Component", in this context, means a major essential component +(kernel, window system, and so on) of the specific operating system +(if any) on which the executable work runs, or a compiler used to +produce the work, or an object code interpreter used to run it. + + The "Corresponding Source" for a work in object code form means all +the source code needed to generate, install, and (for an executable +work) run the object code and to modify the work, including scripts to +control those activities. However, it does not include the work's +System Libraries, or general-purpose tools or generally available free +programs which are used unmodified in performing those activities but +which are not part of the work. For example, Corresponding Source +includes interface definition files associated with source files for +the work, and the source code for shared libraries and dynamically +linked subprograms that the work is specifically designed to require, +such as by intimate data communication or control flow between those +subprograms and other parts of the work. + + The Corresponding Source need not include anything that users +can regenerate automatically from other parts of the Corresponding +Source. + + The Corresponding Source for a work in source code form is that +same work. + + 2. Basic Permissions. + + All rights granted under this License are granted for the term of +copyright on the Program, and are irrevocable provided the stated +conditions are met. This License explicitly affirms your unlimited +permission to run the unmodified Program. The output from running a +covered work is covered by this License only if the output, given its +content, constitutes a covered work. This License acknowledges your +rights of fair use or other equivalent, as provided by copyright law. + + You may make, run and propagate covered works that you do not +convey, without conditions so long as your license otherwise remains +in force. You may convey covered works to others for the sole purpose +of having them make modifications exclusively for you, or provide you +with facilities for running those works, provided that you comply with +the terms of this License in conveying all material for which you do +not control copyright. Those thus making or running the covered works +for you must do so exclusively on your behalf, under your direction +and control, on terms that prohibit them from making any copies of +your copyrighted material outside their relationship with you. + + Conveying under any other circumstances is permitted solely under +the conditions stated below. Sublicensing is not allowed; section 10 +makes it unnecessary. + + 3. Protecting Users' Legal Rights From Anti-Circumvention Law. + + No covered work shall be deemed part of an effective technological +measure under any applicable law fulfilling obligations under article +11 of the WIPO copyright treaty adopted on 20 December 1996, or +similar laws prohibiting or restricting circumvention of such +measures. + + When you convey a covered work, you waive any legal power to forbid +circumvention of technological measures to the extent such circumvention +is effected by exercising rights under this License with respect to +the covered work, and you disclaim any intention to limit operation or +modification of the work as a means of enforcing, against the work's +users, your or third parties' legal rights to forbid circumvention of +technological measures. + + 4. Conveying Verbatim Copies. + + You may convey verbatim copies of the Program's source code as you +receive it, in any medium, provided that you conspicuously and +appropriately publish on each copy an appropriate copyright notice; +keep intact all notices stating that this License and any +non-permissive terms added in accord with section 7 apply to the code; +keep intact all notices of the absence of any warranty; and give all +recipients a copy of this License along with the Program. + + You may charge any price or no price for each copy that you convey, +and you may offer support or warranty protection for a fee. + + 5. Conveying Modified Source Versions. + + You may convey a work based on the Program, or the modifications to +produce it from the Program, in the form of source code under the +terms of section 4, provided that you also meet all of these conditions: + + a) The work must carry prominent notices stating that you modified + it, and giving a relevant date. + + b) The work must carry prominent notices stating that it is + released under this License and any conditions added under section + 7. This requirement modifies the requirement in section 4 to + "keep intact all notices". + + c) You must license the entire work, as a whole, under this + License to anyone who comes into possession of a copy. This + License will therefore apply, along with any applicable section 7 + additional terms, to the whole of the work, and all its parts, + regardless of how they are packaged. This License gives no + permission to license the work in any other way, but it does not + invalidate such permission if you have separately received it. + + d) If the work has interactive user interfaces, each must display + Appropriate Legal Notices; however, if the Program has interactive + interfaces that do not display Appropriate Legal Notices, your + work need not make them do so. + + A compilation of a covered work with other separate and independent +works, which are not by their nature extensions of the covered work, +and which are not combined with it such as to form a larger program, +in or on a volume of a storage or distribution medium, is called an +"aggregate" if the compilation and its resulting copyright are not +used to limit the access or legal rights of the compilation's users +beyond what the individual works permit. Inclusion of a covered work +in an aggregate does not cause this License to apply to the other +parts of the aggregate. + + 6. Conveying Non-Source Forms. + + You may convey a covered work in object code form under the terms +of sections 4 and 5, provided that you also convey the +machine-readable Corresponding Source under the terms of this License, +in one of these ways: + + a) Convey the object code in, or embodied in, a physical product + (including a physical distribution medium), accompanied by the + Corresponding Source fixed on a durable physical medium + customarily used for software interchange. + + b) Convey the object code in, or embodied in, a physical product + (including a physical distribution medium), accompanied by a + written offer, valid for at least three years and valid for as + long as you offer spare parts or customer support for that product + model, to give anyone who possesses the object code either (1) a + copy of the Corresponding Source for all the software in the + product that is covered by this License, on a durable physical + medium customarily used for software interchange, for a price no + more than your reasonable cost of physically performing this + conveying of source, or (2) access to copy the + Corresponding Source from a network server at no charge. + + c) Convey individual copies of the object code with a copy of the + written offer to provide the Corresponding Source. This + alternative is allowed only occasionally and noncommercially, and + only if you received the object code with such an offer, in accord + with subsection 6b. + + d) Convey the object code by offering access from a designated + place (gratis or for a charge), and offer equivalent access to the + Corresponding Source in the same way through the same place at no + further charge. You need not require recipients to copy the + Corresponding Source along with the object code. If the place to + copy the object code is a network server, the Corresponding Source + may be on a different server (operated by you or a third party) + that supports equivalent copying facilities, provided you maintain + clear directions next to the object code saying where to find the + Corresponding Source. Regardless of what server hosts the + Corresponding Source, you remain obligated to ensure that it is + available for as long as needed to satisfy these requirements. + + e) Convey the object code using peer-to-peer transmission, provided + you inform other peers where the object code and Corresponding + Source of the work are being offered to the general public at no + charge under subsection 6d. + + A separable portion of the object code, whose source code is excluded +from the Corresponding Source as a System Library, need not be +included in conveying the object code work. + + A "User Product" is either (1) a "consumer product", which means any +tangible personal property which is normally used for personal, family, +or household purposes, or (2) anything designed or sold for incorporation +into a dwelling. In determining whether a product is a consumer product, +doubtful cases shall be resolved in favor of coverage. For a particular +product received by a particular user, "normally used" refers to a +typical or common use of that class of product, regardless of the status +of the particular user or of the way in which the particular user +actually uses, or expects or is expected to use, the product. A product +is a consumer product regardless of whether the product has substantial +commercial, industrial or non-consumer uses, unless such uses represent +the only significant mode of use of the product. + + "Installation Information" for a User Product means any methods, +procedures, authorization keys, or other information required to install +and execute modified versions of a covered work in that User Product from +a modified version of its Corresponding Source. The information must +suffice to ensure that the continued functioning of the modified object +code is in no case prevented or interfered with solely because +modification has been made. + + If you convey an object code work under this section in, or with, or +specifically for use in, a User Product, and the conveying occurs as +part of a transaction in which the right of possession and use of the +User Product is transferred to the recipient in perpetuity or for a +fixed term (regardless of how the transaction is characterized), the +Corresponding Source conveyed under this section must be accompanied +by the Installation Information. But this requirement does not apply +if neither you nor any third party retains the ability to install +modified object code on the User Product (for example, the work has +been installed in ROM). + + The requirement to provide Installation Information does not include a +requirement to continue to provide support service, warranty, or updates +for a work that has been modified or installed by the recipient, or for +the User Product in which it has been modified or installed. Access to a +network may be denied when the modification itself materially and +adversely affects the operation of the network or violates the rules and +protocols for communication across the network. + + Corresponding Source conveyed, and Installation Information provided, +in accord with this section must be in a format that is publicly +documented (and with an implementation available to the public in +source code form), and must require no special password or key for +unpacking, reading or copying. + + 7. Additional Terms. + + "Additional permissions" are terms that supplement the terms of this +License by making exceptions from one or more of its conditions. +Additional permissions that are applicable to the entire Program shall +be treated as though they were included in this License, to the extent +that they are valid under applicable law. If additional permissions +apply only to part of the Program, that part may be used separately +under those permissions, but the entire Program remains governed by +this License without regard to the additional permissions. + + When you convey a copy of a covered work, you may at your option +remove any additional permissions from that copy, or from any part of +it. (Additional permissions may be written to require their own +removal in certain cases when you modify the work.) You may place +additional permissions on material, added by you to a covered work, +for which you have or can give appropriate copyright permission. + + Notwithstanding any other provision of this License, for material you +add to a covered work, you may (if authorized by the copyright holders of +that material) supplement the terms of this License with terms: + + a) Disclaiming warranty or limiting liability differently from the + terms of sections 15 and 16 of this License; or + + b) Requiring preservation of specified reasonable legal notices or + author attributions in that material or in the Appropriate Legal + Notices displayed by works containing it; or + + c) Prohibiting misrepresentation of the origin of that material, or + requiring that modified versions of such material be marked in + reasonable ways as different from the original version; or + + d) Limiting the use for publicity purposes of names of licensors or + authors of the material; or + + e) Declining to grant rights under trademark law for use of some + trade names, trademarks, or service marks; or + + f) Requiring indemnification of licensors and authors of that + material by anyone who conveys the material (or modified versions of + it) with contractual assumptions of liability to the recipient, for + any liability that these contractual assumptions directly impose on + those licensors and authors. + + All other non-permissive additional terms are considered "further +restrictions" within the meaning of section 10. If the Program as you +received it, or any part of it, contains a notice stating that it is +governed by this License along with a term that is a further +restriction, you may remove that term. If a license document contains +a further restriction but permits relicensing or conveying under this +License, you may add to a covered work material governed by the terms +of that license document, provided that the further restriction does +not survive such relicensing or conveying. + + If you add terms to a covered work in accord with this section, you +must place, in the relevant source files, a statement of the +additional terms that apply to those files, or a notice indicating +where to find the applicable terms. + + Additional terms, permissive or non-permissive, may be stated in the +form of a separately written license, or stated as exceptions; +the above requirements apply either way. + + 8. Termination. + + You may not propagate or modify a covered work except as expressly +provided under this License. Any attempt otherwise to propagate or +modify it is void, and will automatically terminate your rights under +this License (including any patent licenses granted under the third +paragraph of section 11). + + However, if you cease all violation of this License, then your +license from a particular copyright holder is reinstated (a) +provisionally, unless and until the copyright holder explicitly and +finally terminates your license, and (b) permanently, if the copyright +holder fails to notify you of the violation by some reasonable means +prior to 60 days after the cessation. + + Moreover, your license from a particular copyright holder is +reinstated permanently if the copyright holder notifies you of the +violation by some reasonable means, this is the first time you have +received notice of violation of this License (for any work) from that +copyright holder, and you cure the violation prior to 30 days after +your receipt of the notice. + + Termination of your rights under this section does not terminate the +licenses of parties who have received copies or rights from you under +this License. If your rights have been terminated and not permanently +reinstated, you do not qualify to receive new licenses for the same +material under section 10. + + 9. Acceptance Not Required for Having Copies. + + You are not required to accept this License in order to receive or +run a copy of the Program. Ancillary propagation of a covered work +occurring solely as a consequence of using peer-to-peer transmission +to receive a copy likewise does not require acceptance. However, +nothing other than this License grants you permission to propagate or +modify any covered work. These actions infringe copyright if you do +not accept this License. Therefore, by modifying or propagating a +covered work, you indicate your acceptance of this License to do so. + + 10. Automatic Licensing of Downstream Recipients. + + Each time you convey a covered work, the recipient automatically +receives a license from the original licensors, to run, modify and +propagate that work, subject to this License. You are not responsible +for enforcing compliance by third parties with this License. + + An "entity transaction" is a transaction transferring control of an +organization, or substantially all assets of one, or subdividing an +organization, or merging organizations. If propagation of a covered +work results from an entity transaction, each party to that +transaction who receives a copy of the work also receives whatever +licenses to the work the party's predecessor in interest had or could +give under the previous paragraph, plus a right to possession of the +Corresponding Source of the work from the predecessor in interest, if +the predecessor has it or can get it with reasonable efforts. + + You may not impose any further restrictions on the exercise of the +rights granted or affirmed under this License. For example, you may +not impose a license fee, royalty, or other charge for exercise of +rights granted under this License, and you may not initiate litigation +(including a cross-claim or counterclaim in a lawsuit) alleging that +any patent claim is infringed by making, using, selling, offering for +sale, or importing the Program or any portion of it. + + 11. Patents. + + A "contributor" is a copyright holder who authorizes use under this +License of the Program or a work on which the Program is based. The +work thus licensed is called the contributor's "contributor version". + + A contributor's "essential patent claims" are all patent claims +owned or controlled by the contributor, whether already acquired or +hereafter acquired, that would be infringed by some manner, permitted +by this License, of making, using, or selling its contributor version, +but do not include claims that would be infringed only as a +consequence of further modification of the contributor version. For +purposes of this definition, "control" includes the right to grant +patent sublicenses in a manner consistent with the requirements of +this License. + + Each contributor grants you a non-exclusive, worldwide, royalty-free +patent license under the contributor's essential patent claims, to +make, use, sell, offer for sale, import and otherwise run, modify and +propagate the contents of its contributor version. + + In the following three paragraphs, a "patent license" is any express +agreement or commitment, however denominated, not to enforce a patent +(such as an express permission to practice a patent or covenant not to +sue for patent infringement). To "grant" such a patent license to a +party means to make such an agreement or commitment not to enforce a +patent against the party. + + If you convey a covered work, knowingly relying on a patent license, +and the Corresponding Source of the work is not available for anyone +to copy, free of charge and under the terms of this License, through a +publicly available network server or other readily accessible means, +then you must either (1) cause the Corresponding Source to be so +available, or (2) arrange to deprive yourself of the benefit of the +patent license for this particular work, or (3) arrange, in a manner +consistent with the requirements of this License, to extend the patent +license to downstream recipients. "Knowingly relying" means you have +actual knowledge that, but for the patent license, your conveying the +covered work in a country, or your recipient's use of the covered work +in a country, would infringe one or more identifiable patents in that +country that you have reason to believe are valid. + + If, pursuant to or in connection with a single transaction or +arrangement, you convey, or propagate by procuring conveyance of, a +covered work, and grant a patent license to some of the parties +receiving the covered work authorizing them to use, propagate, modify +or convey a specific copy of the covered work, then the patent license +you grant is automatically extended to all recipients of the covered +work and works based on it. + + A patent license is "discriminatory" if it does not include within +the scope of its coverage, prohibits the exercise of, or is +conditioned on the non-exercise of one or more of the rights that are +specifically granted under this License. You may not convey a covered +work if you are a party to an arrangement with a third party that is +in the business of distributing software, under which you make payment +to the third party based on the extent of your activity of conveying +the work, and under which the third party grants, to any of the +parties who would receive the covered work from you, a discriminatory +patent license (a) in connection with copies of the covered work +conveyed by you (or copies made from those copies), or (b) primarily +for and in connection with specific products or compilations that +contain the covered work, unless you entered into that arrangement, +or that patent license was granted, prior to 28 March 2007. + + Nothing in this License shall be construed as excluding or limiting +any implied license or other defenses to infringement that may +otherwise be available to you under applicable patent law. + + 12. No Surrender of Others' Freedom. + + If conditions are imposed on you (whether by court order, agreement or +otherwise) that contradict the conditions of this License, they do not +excuse you from the conditions of this License. If you cannot convey a +covered work so as to satisfy simultaneously your obligations under this +License and any other pertinent obligations, then as a consequence you may +not convey it at all. For example, if you agree to terms that obligate you +to collect a royalty for further conveying from those to whom you convey +the Program, the only way you could satisfy both those terms and this +License would be to refrain entirely from conveying the Program. + + 13. Use with the GNU Affero General Public License. + + Notwithstanding any other provision of this License, you have +permission to link or combine any covered work with a work licensed +under version 3 of the GNU Affero General Public License into a single +combined work, and to convey the resulting work. The terms of this +License will continue to apply to the part which is the covered work, +but the special requirements of the GNU Affero General Public License, +section 13, concerning interaction through a network will apply to the +combination as such. + + 14. Revised Versions of this License. + + The Free Software Foundation may publish revised and/or new versions of +the GNU General Public License from time to time. Such new versions will +be similar in spirit to the present version, but may differ in detail to +address new problems or concerns. + + Each version is given a distinguishing version number. If the +Program specifies that a certain numbered version of the GNU General +Public License "or any later version" applies to it, you have the +option of following the terms and conditions either of that numbered +version or of any later version published by the Free Software +Foundation. If the Program does not specify a version number of the +GNU General Public License, you may choose any version ever published +by the Free Software Foundation. + + If the Program specifies that a proxy can decide which future +versions of the GNU General Public License can be used, that proxy's +public statement of acceptance of a version permanently authorizes you +to choose that version for the Program. + + Later license versions may give you additional or different +permissions. However, no additional obligations are imposed on any +author or copyright holder as a result of your choosing to follow a +later version. + + 15. Disclaimer of Warranty. + + THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY +APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT +HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY +OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, +THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR +PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM +IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF +ALL NECESSARY SERVICING, REPAIR OR CORRECTION. + + 16. Limitation of Liability. + + IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING +WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS +THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY +GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE +USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF +DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD +PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS), +EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF +SUCH DAMAGES. + + 17. Interpretation of Sections 15 and 16. + + If the disclaimer of warranty and limitation of liability provided +above cannot be given local legal effect according to their terms, +reviewing courts shall apply local law that most closely approximates +an absolute waiver of all civil liability in connection with the +Program, unless a warranty or assumption of liability accompanies a +copy of the Program in return for a fee. + + END OF TERMS AND CONDITIONS + + How to Apply These Terms to Your New Programs + + If you develop a new program, and you want it to be of the greatest +possible use to the public, the best way to achieve this is to make it +free software which everyone can redistribute and change under these terms. + + To do so, attach the following notices to the program. It is safest +to attach them to the start of each source file to most effectively +state the exclusion of warranty; and each file should have at least +the "copyright" line and a pointer to where the full notice is found. + + + Copyright (C) + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + +Also add information on how to contact you by electronic and paper mail. + + If the program does terminal interaction, make it output a short +notice like this when it starts in an interactive mode: + + Copyright (C) + This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'. + This is free software, and you are welcome to redistribute it + under certain conditions; type `show c' for details. + +The hypothetical commands `show w' and `show c' should show the appropriate +parts of the General Public License. Of course, your program's commands +might be different; for a GUI interface, you would use an "about box". + + You should also get your employer (if you work as a programmer) or school, +if any, to sign a "copyright disclaimer" for the program, if necessary. +For more information on this, and how to apply and follow the GNU GPL, see +. + + The GNU General Public License does not permit incorporating your program +into proprietary programs. If your program is a subroutine library, you +may consider it more useful to permit linking proprietary applications with +the library. If this is what you want to do, use the GNU Lesser General +Public License instead of this License. But first, please read +. diff --git a/README.md b/README.md new file mode 100644 index 0000000..8ae2102 --- /dev/null +++ b/README.md @@ -0,0 +1,60 @@ +# GelSight SDK +[![License: GPL v3](https://img.shields.io/badge/License-GPLv3-blue.svg)](https://www.gnu.org/licenses/gpl-3.0)  [](https://rpl.ri.cmu.edu/) + +This repository is a modified version of the [the official gsrobotics implementation](https://github.com/gelsightinc/gsrobotics), offering improvements in usability and sensor compatibility. + +## Key Features + +- **Sensor Calibration**: Added functionality for users to calibrate their own GelSight sensors. + +- **Expanded Sensor Compatibility**: Compatible with other GelSight sensors, including Digit and lab-made sensors. + +- **Low-Latency Sensor Reading**: Including an enhanced image streaming pipeline for reduced latency and frame drop, especially for GelSight Mini. + +Authors: +* [Hung-Jui Huang](https://joehjhuang.github.io/) (hungjuih@andrew.cmu.edu) +* Ruihan Gao (ruihang@andrew.cmu.edu) + +## Support System +* Tested on Ubuntu 22.04 +* Tested on GelSight Mini and Digit +* Python >= 3.9 + +## Installation +Clone and install gs_sdk from source: +```bash +git clone git@github.com:joehjhuang/gs_sdk.git +cd gs_sdk +pip install -e . +``` + +## Coordinate Conventions +The coordinate system convention in this SDK is shown below, using the GelSight Mini sensor for illustration: + +| 2D (sensor image) | 3D | +| --------------------------------- | --------------------------------- | +| | | + +## Sensor Calibration +For more details on sensor calibration, see the [Calibration README](calibration/README.md). + +## Examples +These examples show basic usage of this GelSight SDK. +### Sensor Streaming +Stream images from a connected GelSight Mini: +```python +python examples/stream_device.py +``` + +### Low Latency Sensor Streaming +Stream images with low latency and without frame dropping from a connected GelSight Mini: +```python +python examples/fast_stream_device.py +``` + +### Reconstruct Touched Surface +Reconstruct a touched surface using the calibration model. Calibration steps are detailed in the [Calibration README](calibration/README.md). +```python +python examples/reconstruct.py +``` +The reconstructed surface will be displayed and saved in `examples/data`. \ No newline at end of file diff --git a/assets/ball_image.png b/assets/ball_image.png new file mode 100644 index 0000000..28760ad Binary files /dev/null and b/assets/ball_image.png differ diff --git a/assets/ball_indenter.jpg b/assets/ball_indenter.jpg new file mode 100644 index 0000000..147bf54 Binary files /dev/null and b/assets/ball_indenter.jpg differ diff --git a/assets/gsmini_frame_2D.png b/assets/gsmini_frame_2D.png new file mode 100644 index 0000000..290743b Binary files /dev/null and b/assets/gsmini_frame_2D.png differ diff --git a/assets/gsmini_frame_3D.png b/assets/gsmini_frame_3D.png new file mode 100644 index 0000000..822221c Binary files /dev/null and b/assets/gsmini_frame_3D.png differ diff --git a/assets/nanogui.png b/assets/nanogui.png new file mode 100644 index 0000000..92e2de3 Binary files /dev/null and b/assets/nanogui.png differ diff --git a/assets/pressing.jpg b/assets/pressing.jpg new file mode 100644 index 0000000..a694d13 Binary files /dev/null and b/assets/pressing.jpg differ diff --git a/assets/rpl.png b/assets/rpl.png new file mode 100644 index 0000000..962fe64 Binary files /dev/null and b/assets/rpl.png differ diff --git a/calibration/README.md b/calibration/README.md new file mode 100644 index 0000000..b5fe019 --- /dev/null +++ b/calibration/README.md @@ -0,0 +1,68 @@ +## Sensor Calibration +We provide calibration tools for new GelSight sensors, following the method described in [1]; the process generally takes less than an hour. For lab-made sensors, refer to `examples/configs/gsmini.yaml` to create a sensor specification, which will be used as the `CONFIG_PATH` in the command below. Below are step-by-step instructions for sensor calibration. **Note: The current version does not support gel pads with markers.** +### Calibration Data Collection +| Ball Indenter | Collecting Data | Collected Image | +|---------|---------|---------| +| | | | + +To collect calibration data, use a ball indenter of known diameter to press against the sensor. Examples of the setup and resulting images are shown above. Run the following command and allocate a location `CALIB_DIR` to save the calibration data: +```bash +collect_data [-b CALIB_DIR] [-d BALL_DIAMETER_IN_MM] [-c CONFIG_PATH] +``` +* Instruction: + * Save background image: Press 'b' + * Capture ~50 tactile images by pressing the ball in various locations: Press 'w' + * Exit: Press 'q' +* Tips for Optimal Calibration: + * Ball Size: Select a ball that appears well-sized within the sensor’s view, like the tactile image shown above; 4mm to 9mm is suitable for GelSight Mini. + * Pressure: Avoid pressing too hard. + * Coverage: Capture at least 50 images with the ball contacting different regions of the sensor. + * Using Multiple Balls: Use the same `CALIB_DIR` and specify distinct `BALL_DIAMETER` values if balls in different size are applied. + +### Label Collected Data +| NanoGui Screenshot | +|---------| +| | + +Run the command below to label the contact circle on the collected tactile data using NanoGUI: + +```bash +label_data [-b CALIB_DIR] [-c CONFIG_PATH] +``` +* Instruction: + * Click the **Open** icon to begin. + * Keyboard Controls for aligning the label with the contact circle: + * **Arrow keys (left/right/up/down)**: Adjust the circle's position. + * **'m' / 'p'**: Decrease / increase the circle's radius. + * **'f' / 'c'**: Decrease / increase the circle's movement step. + * Once aligned, click the **Calibrate** icon. + * After labeling all data, close the window to exit. + +### Prepare Dataset +Run the command below to prepare the dataset for calibration model training: +```bash +prepare_data [-b CALIB_DIR] [-c CONFIG_PATH] +``` + +### Train Calibration Model +Train the MLP model to map pixel color and location (RGBXY) to surface gradients for each pixel. Use the following command to train the model with the collected dataset: +```bash +train_model [-b CALIB_DIR] [-d {cpu|cuda}] +``` + +The trained model is saved in `CALIB_DIR/model/nnmodel.pth`. + +## Test the Trained Calibration Model +Once the model is trained, connect the sensor and run the following command to stream images and perform real-time surface reconstruction using the trained calibration model: + +```bash +test_model [-b CALIB_DIR] [-c CONFIG_PATH] +``` +After starting, wait briefly for background data collection; real-time surface gradient predictions will then be displayed. Press any key to exit. + + +### References +1. S. Wang, Y. She, B. Romero, and E. H. Adelson, “Gelsight wedge: +Measuring high-resolution 3d contact geometry with a compact robot +finger,” in 2021 IEEE International Conference on Robotics and +Automation (ICRA). IEEE, 2021. \ No newline at end of file diff --git a/calibration/__init__.py b/calibration/__init__.py new file mode 100644 index 0000000..e69de29 diff --git a/calibration/collect_data.py b/calibration/collect_data.py new file mode 100644 index 0000000..46ae7df --- /dev/null +++ b/calibration/collect_data.py @@ -0,0 +1,148 @@ +import argparse +import os + +import cv2 +import numpy as np +import yaml + +from gs_sdk.gs_device import Camera +from calibration.utils import load_csv_as_dict + +""" +This script collects tactile data using ball indenters for sensor calibration. + +Instruction: + 1. Connect the sensor to the computer. + 2. Prepare a ball indenter with known diameter. + 3. Runs this script, press 'b' to collect a background image. + 4. Press the sensor with the ball indenter at multiple locations (~50 locations preferred), + press 'w' to save the tactile image. When done, press 'q' to quit. +Note: + If you have prepared multiple balls in different diameters, you can run this script multiple + times, assign the same calib_dir but different ball diameters, the system will treat it as + one single dataset. + +Usage: + python collect_data.py --calib_dir CALIB_DIR --ball_diameter DIAMETER [--config_path CONFIG_PATH] + +Arguments: + --calib_dir: Path to the directory where the collected data will be saved + --ball_diameter: Diameter of the ball indenter in mm + --config_path: (Optional) Path to the configuration file about the sensor dimensions. + If not provided, GelSight Mini is assumed. +""" + +config_dir = os.path.join(os.path.dirname(__file__), "../examples/configs") + + +def collect_data(): + # Argument Parsers + parser = argparse.ArgumentParser( + description="Collect calibration data with ball indenters to calibrate the sensor." + ) + parser.add_argument( + "-b", + "--calib_dir", + type=str, + help="path to save calibration data", + ) + parser.add_argument( + "-d", "--ball_diameter", type=float, help="diameter of the indenter in mm" + ) + parser.add_argument( + "-c", + "--config_path", + type=str, + help="path of the sensor information", + default=os.path.join(config_dir, "gsmini.yaml"), + ) + args = parser.parse_args() + + # Create the data saving directories + calib_dir = args.calib_dir + ball_diameter = args.ball_diameter + indenter_subdir = "%.3fmm" % (ball_diameter) + indenter_dir = os.path.join(calib_dir, indenter_subdir) + if not os.path.isdir(indenter_dir): + os.makedirs(indenter_dir) + + # Read the configuration + config_path = args.config_path + with open(config_path, "r") as f: + config = yaml.safe_load(f) + device_name = config["device_name"] + imgh = config["imgh"] + imgw = config["imgw"] + + # Create the data saving catalog + catalog_path = os.path.join(calib_dir, "catalog.csv") + if not os.path.isfile(catalog_path): + with open(catalog_path, "w") as f: + f.write("experiment_reldir,diameter(mm)\n") + + # Find last data_count collected with this diameter + data_dict = load_csv_as_dict(catalog_path) + diameters = np.array([float(diameter) for diameter in data_dict["diameter(mm)"]]) + data_idxs = np.where(np.abs(diameters - ball_diameter) < 1e-3)[0] + data_counts = np.array( + [int(os.path.basename(reldir)) for reldir in data_dict["experiment_reldir"]] + ) + if len(data_idxs) == 0: + data_count = 0 + else: + data_count = max(data_counts[data_idxs]) + 1 + + # Connect to the device and collect data until quit + device = Camera(device_name, imgh, imgw) + device.connect() + print("Press key to collect data, collect background, or quit (w/b/q)") + while True: + image = device.get_image() + + # Display the image and decide record or quit + cv2.imshow("frame", image) + key = cv2.waitKey(100) + if key == ord("w"): + # Save the image + experiment_reldir = os.path.join(indenter_subdir, str(data_count)) + experiment_dir = os.path.join(calib_dir, experiment_reldir) + if not os.path.isdir(experiment_dir): + os.makedirs(experiment_dir) + save_path = os.path.join(experiment_dir, "gelsight.png") + cv2.imwrite(save_path, image) + print("Save data to new path: %s" % save_path) + + # Save to catalog + with open(catalog_path, "a") as f: + f.write(experiment_reldir + "," + str(ball_diameter)) + f.write("\n") + data_count += 1 + elif key == ord("b"): + print("Collecting 10 background images, please wait ...") + images = [] + for _ in range(10): + image = device.get_image() + images.append(image) + cv2.imshow("frame", image) + cv2.waitKey(1) + image = np.mean(images, axis=0).astype(np.uint8) + # Save the background image + save_path = os.path.join(calib_dir, "background.png") + cv2.imwrite(save_path, image) + print("Save background image to %s" % save_path) + elif key == ord("q"): + # Quit + break + elif key == -1: + # No key pressed + continue + else: + print("Unrecognized key %s" % key) + + device.release() + cv2.destroyAllWindows() + print("%d images collected in total." % data_count) + + +if __name__ == "__main__": + collect_data() diff --git a/calibration/label_data.py b/calibration/label_data.py new file mode 100644 index 0000000..eb88c67 --- /dev/null +++ b/calibration/label_data.py @@ -0,0 +1,314 @@ +import gc +import os +import argparse + +import cv2 +import numpy as np +import nanogui as ng +from nanogui import Texture +from nanogui import glfw +import yaml + +from calibration.utils import load_csv_as_dict + +""" +Rewrite from Zilin Si's code: https://github.com/Robo-Touch/Taxim + +This script is for manually labeling the contact circle in the tactile image for sensor calibration. +The labeled data will save the center and radius of the circle in the tactile image. + +Prerequisite: + - Tactile images collected using ball indenters with known diameters are collected. +Instruction: + 1. Runs the script, + 2. Mouse-press 'Open' to select the directory where the collected data are saved. + 3. Press left/right/up/down to control the circle's location, + Press m/p to decrease/increase the circle's radius, + Press f/c to decrease/increase the circle's moving step. + Mouse-press 'Calibrate' to save the labeled data. + 4. Repeat step 3 for all the tactile images and close Nanogui when done. + +Usage: + python label_data.py --calib_dir CALIB_DIR [--config_path CONFIG_PATH] [--display_difference] [--detect_circle] + +Arguments: + --calib_dir: Path to the directory where the collected data are saved + --config_path: (Optional) Path to the configuration file about the sensor dimensions. + If not provided, GelSight Mini is assumed. + --display_difference: (Store True) Display the difference between the background image. + --detect_circle: (Store True) Automatically detect the circle in the image. +""" + +config_dir = os.path.join(os.path.dirname(__file__), "../examples/configs") + + +class Circle: + """the circle drawed on the tactile image to get the contact size""" + + color_circle = (128, 0, 0) + opacity = 0.5 + + def __init__(self, x, y, radius=25, increments=2): + self.center = [x, y] + self.radius = radius + self.increments = increments + + +class CalibrateApp(ng.Screen): + fnames = list() + read_all = False # flag to indicate if all images have been read + load_img = True + change = False + + def __init__( + self, calib_data, imgw, imgh, display_difference=False, detect_circle=False + ): + super(CalibrateApp, self).__init__((1024, 768), "Gelsight Calibration App") + self.imgw = imgw + self.imgh = imgh + self.display_difference = display_difference + self.detect_circle = detect_circle + # Load background + self.bg_img = cv2.imread(os.path.join(calib_data, "background.png")) + # Initialize the circle + self.circle = Circle(self.imgw / 2, self.imgh / 2, radius=40) + + window = ng.Window(self, "IO Window") + window.set_position((15, 15)) + window.set_layout(ng.GroupLayout()) + + ng.Label(window, "Folder dialog", "sans-bold") + tools = ng.Widget(window) + tools.set_layout( + ng.BoxLayout(ng.Orientation.Horizontal, ng.Alignment.Middle, 0, 6) + ) + + # Initialize the file directory and list of filenames + b = ng.Button(tools, "Open") + + def open_cb(): + self.parent_dir = calib_data + # Read the catalog and create a list of all filenames + catalog_dict = load_csv_as_dict( + os.path.join(self.parent_dir, "catalog.csv") + ) + self.fnames = [ + os.path.join(self.parent_dir, fname) + for fname in catalog_dict["experiment_reldir"] + ] + self.circle_radius = [ + float(radius) for radius in catalog_dict["diameter(mm)"] + ] + print( + f"Selected directory = {self.parent_dir}, total {len(self.fnames)} images" + ) + self.img_idx = 0 + + b.set_callback(open_cb) + + # Initialize the image window + self.img_window = ng.Window(self, "Current image") + self.img_window.set_position((200, 15)) + self.img_window.set_layout(ng.GroupLayout()) + + # Initialize the calibrate button + b = ng.Button(self.img_window, "Calibrate") + + def calibrate_cb(): + frame = self.orig_img + center = self.circle.center + radius = self.circle.radius + print(f"Frame {self.img_idx}: center = {center}, radius = is {radius}") + # save the data for each individual frame instead of creating a long list and save it at the end + # save the radius and center to npz file + save_dir = os.path.join(self.fnames[self.img_idx], "label.npz") + np.savez(save_dir, center=center, radius=radius) + # save the labeled image + labeled_img = self.overlay_circle(frame, self.circle) + labeled_img_path = os.path.join(self.fnames[self.img_idx], "labeled.png") + cv2.imwrite(labeled_img_path, labeled_img) + + # Update img index + self.load_img = True + self.update_img_idx() + + b.set_callback(calibrate_cb) + + ########### + self.img_view = ng.ImageView(self.img_window) + self.img_tex = ng.Texture( + pixel_format=Texture.PixelFormat.RGB, + component_format=Texture.ComponentFormat.UInt8, + size=[imgw, imgh], + min_interpolation_mode=Texture.InterpolationMode.Trilinear, + mag_interpolation_mode=Texture.InterpolationMode.Nearest, + flags=Texture.TextureFlags.ShaderRead | Texture.TextureFlags.RenderTarget, + ) + self.perform_layout() + + def update_img_idx(self): + self.img_idx += 1 + if self.img_idx == len(self.fnames) - 1: + self.read_all = True + + def overlay_circle(self, orig_img, circle): + center = circle.center + radius = circle.radius + color_circle = circle.color_circle + opacity = circle.opacity + + overlay = orig_img.copy() + center_tuple = (int(center[0]), int(center[1])) + cv2.circle(overlay, center_tuple, radius, color_circle, -1) + cv2.addWeighted(overlay, opacity, orig_img, 1 - opacity, 0, overlay) + return overlay + + def draw(self, ctx): + self.img_window.set_size((2000, 2600)) + self.img_view.set_size((self.imgw, self.imgh)) + + # load a new image + if self.load_img and len(self.fnames) > 0 and not self.read_all: + print("Loading %s" % self.fnames[self.img_idx]) + + # Load img + self.orig_img = cv2.imread( + os.path.join(self.fnames[self.img_idx], "gelsight.png") + ) + # Initialize the circle pose + if self.detect_circle: + diff_image = self.orig_img.astype(np.float32) - self.bg_img.astype( + np.float32 + ) + color_mask = np.linalg.norm(diff_image, axis=-1) > 15 + color_mask = cv2.dilate( + color_mask.astype(np.uint8), np.ones((7, 7), np.uint8) + ) + color_mask = cv2.erode( + color_mask.astype(np.uint8), np.ones((15, 15), np.uint8) + ) + contours, _ = cv2.findContours( + color_mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE + ) + largest_contour = max(contours, key=cv2.contourArea) + M = cv2.moments(largest_contour) + cx = int(M["m10"] / M["m00"]) + cy = int(M["m01"] / M["m00"]) + else: + cx = self.circle.center[0] + cy = self.circle.center[1] + radius = max(self.circle.radius - 13, 5) + self.circle = Circle(cx, cy, radius=radius) + + # Add circle and add img to viewer + if (self.load_img and len(self.fnames) > 0) or self.change: + self.load_img = False + self.change = False + # Add circle + if self.display_difference: + diff_img = ( + self.orig_img.astype(np.float32) - self.bg_img.astype(np.float32) + ) * 3 + diff_img = np.clip(diff_img, -127, 128) + np.ones_like(diff_img) * 127 + display_img = cv2.cvtColor(diff_img.astype(np.uint8), cv2.COLOR_BGR2RGB) + else: + display_img = cv2.cvtColor(self.orig_img, cv2.COLOR_BGR2RGB) + img = self.overlay_circle(display_img, self.circle) + + if self.img_tex.channels() > 3: + height, width = img.shape[:2] + alpha = 255 * np.ones((height, width, 1), dtype=img.dtype) + img = np.concatenate((img, alpha), axis=2) + + # Add to img view + self.img_tex.upload(img) + self.img_view.set_image(self.img_tex) + + super(CalibrateApp, self).draw(ctx) + + def keyboard_event(self, key, scancode, action, modifiers): + if super(CalibrateApp, self).keyboard_event(key, scancode, action, modifiers): + return True + if key == glfw.KEY_ESCAPE and action == glfw.PRESS: + self.set_visible(False) + return True + elif key == glfw.KEY_C: + self.circle.increments *= 2 + elif key == glfw.KEY_F: + self.circle.increments /= 2 + else: + self.change = True + if key == glfw.KEY_LEFT: + self.circle.center[0] -= self.circle.increments + elif key == glfw.KEY_RIGHT: + self.circle.center[0] += self.circle.increments + elif key == glfw.KEY_UP: + self.circle.center[1] -= self.circle.increments + elif key == glfw.KEY_DOWN: + self.circle.center[1] += self.circle.increments + elif key == glfw.KEY_M: + self.circle.radius -= 1 + elif key == glfw.KEY_P: + self.circle.radius += 1 + + return False + + +def label_data(): + # Argument parser + parser = argparse.ArgumentParser( + description="Label the ball indenter data using Nanogui." + ) + parser.add_argument( + "-b", + "--calib_dir", + type=str, + help="path to save calibration data", + ) + parser.add_argument( + "-c", + "--config_path", + type=str, + help="path of configuring gelsight", + default=os.path.join(config_dir, "gsmini.yaml"), + ) + parser.add_argument( + "-d", + "--display_difference", + action="store_true", + help="Display the difference between the background image", + ) + parser.add_argument( + "-r", + "--detect_circle", + action="store_true", + help="Automatically detect the circle in the image", + ) + args = parser.parse_args() + + # Read the configuration + config_path = args.config_path + with open(config_path, "r") as f: + config = yaml.safe_load(f) + imgh = config["imgh"] + imgw = config["imgw"] + + # Start the label process + ng.init() + app = CalibrateApp( + args.calib_dir, + imgw, + imgh, + display_difference=args.display_difference, + detect_circle=args.detect_circle, + ) + app.draw_all() + app.set_visible(True) + ng.mainloop(refresh=1 / 60.0 * 1000) + del app + gc.collect() + ng.shutdown() + + +if __name__ == "__main__": + label_data() diff --git a/calibration/models.py b/calibration/models.py new file mode 100644 index 0000000..cda5356 --- /dev/null +++ b/calibration/models.py @@ -0,0 +1,45 @@ +import torch +import torch.nn as nn +import torch.nn.functional as F +from torch.utils.data import Dataset + + +class BGRXYDataset(Dataset): + """The BGRXY Datast.""" + + def __init__(self, bgrxys, gxyangles): + self.bgrxys = bgrxys + self.gxyangles = gxyangles + + def __len__(self): + return len(self.bgrxys) + + def __getitem__(self, index): + bgrxy = torch.tensor(self.bgrxys[index], dtype=torch.float32) + gxyangles = torch.tensor(self.gxyangles[index], dtype=torch.float32) + return bgrxy, gxyangles + + +class BGRXYMLPNet_(nn.Module): + """ + The architecture using MLP, this is never used in test time. + We train with this architecture and then transfer weights to the 1-by-1 convolution architecture. + """ + + def __init__(self): + super(BGRXYMLPNet_, self).__init__() + input_size = 5 + self.fc1 = nn.Linear(input_size, 128) + self.bn1 = nn.BatchNorm1d(128) + self.fc2 = nn.Linear(128, 32) + self.bn2 = nn.BatchNorm1d(32) + self.fc3 = nn.Linear(32, 32) + self.bn3 = nn.BatchNorm1d(32) + self.fc4 = nn.Linear(32, 2) + + def forward(self, x): + x = F.relu(self.bn1(self.fc1(x))) + x = F.relu(self.bn2(self.fc2(x))) + x = F.relu(self.bn3(self.fc3(x))) + x = self.fc4(x) + return x diff --git a/calibration/prepare_data.py b/calibration/prepare_data.py new file mode 100644 index 0000000..b20c1a1 --- /dev/null +++ b/calibration/prepare_data.py @@ -0,0 +1,132 @@ +import argparse +import json +import os + +import cv2 +import numpy as np +import yaml + +from calibration.utils import load_csv_as_dict +from gs_sdk.gs_reconstruct import image2bgrxys + +""" +This script prepares dataset for the tactile sensor calibration. +It is based on the collected and labeled data. + +Prerequisite: + - Tactile images collected using ball indenters with known diameters are collected. + - Collected tactile images are labeled. + +Usage: + python prepare_data.py --calib_dir CALIB_DIR [--config_path CONFIG_PATH] [--radius_reduction RADIUS_REDUCTION] + +Arguments: + --calib_dir: Path to the directory where the collected data will be saved + --config_path: (Optional) Path to the configuration file about the sensor dimensions. + If not provided, GelSight Mini is assumed. + --radius_reduction: (Optional) Reduce the radius of the labeled circle. This helps guarantee all labeled pixels are indented. + If not provided, 4 pixels will be reduced. +""" + +config_dir = os.path.join(os.path.dirname(__file__), "../examples/configs") + + +def prepare_data(): + # Argument Parsers + parser = argparse.ArgumentParser( + description="Use the labeled collected data to prepare the dataset files (npz)." + ) + parser.add_argument( + "-b", + "--calib_dir", + type=str, + help="path of the calibration data", + ) + parser.add_argument( + "-c", + "--config_path", + type=str, + help="path of configuring gelsight", + default=os.path.join(config_dir, "gsmini.yaml"), + ) + parser.add_argument( + "-r", + "--radius_reduction", + type=float, + help="reduce the radius of the labeled circle. When not considering shadows, this helps guarantee all labeled pixels are indented. ", + default=4.0, + ) + args = parser.parse_args() + + # Load the data_dict + calib_dir = args.calib_dir + catalog_path = os.path.join(calib_dir, "catalog.csv") + data_dict = load_csv_as_dict(catalog_path) + diameters = np.array([float(diameter) for diameter in data_dict["diameter(mm)"]]) + experiment_reldirs = np.array(data_dict["experiment_reldir"]) + + # Split data into train and test and save the split information + perm = np.random.permutation(len(experiment_reldirs)) + n_train = 4 * len(experiment_reldirs) // 5 + data_path = os.path.join(calib_dir, "train_test_split.json") + dict_to_save = { + "train": experiment_reldirs[perm[:n_train]].tolist(), + "test": experiment_reldirs[perm[n_train:]].tolist(), + } + with open(data_path, "w") as f: + json.dump(dict_to_save, f, indent=4) + + # Read the configuration + config_path = args.config_path + with open(config_path, "r") as f: + config = yaml.safe_load(f) + ppmm = config["ppmm"] + + # Extract the pixel data from each tactile image and calculate the gradients + for experiment_reldir, diameter in zip(experiment_reldirs, diameters): + experiment_dir = os.path.join(calib_dir, experiment_reldir) + image_path = os.path.join(experiment_dir, "gelsight.png") + image = cv2.imread(image_path) + + # Filter the non-indented pixels + label_path = os.path.join(experiment_dir, "label.npz") + label_data = np.load(label_path) + center = label_data["center"] + radius = label_data["radius"] - args.radius_reduction + xys = np.dstack( + np.meshgrid( + np.arange(image.shape[1]), np.arange(image.shape[0]), indexing="xy" + ) + ) + dists = np.linalg.norm(xys - center, axis=2) + mask = dists < radius + + # Find the gradient angles, prepare the data, and save the data + ball_radius = diameter / ppmm / 2.0 + if ball_radius < radius: + print(experiment_reldir) + print("Press too deep, deeper than the ball radius") + continue + dxys = xys - center + dists[np.logical_not(mask)] = 0.0 + dzs = np.sqrt(ball_radius**2 - np.square(dists)) + gxangles = np.arctan2(dxys[:, :, 0], dzs) + gyangles = np.arctan2(dxys[:, :, 1], dzs) + gxyangles = np.stack([gxangles, gyangles], axis=-1) + gxyangles[np.logical_not(mask)] = np.array([0.0, 0.0]) + bgrxys = image2bgrxys(image) + save_path = os.path.join(experiment_dir, "data.npz") + np.savez(save_path, bgrxys=bgrxys, gxyangles=gxyangles, mask=mask) + + # Save the background data + bg_path = os.path.join(calib_dir, "background.png") + bg_image = cv2.imread(bg_path) + bgrxys = image2bgrxys(bg_image) + gxyangles = np.zeros((bg_image.shape[0], bg_image.shape[1], 2)) + mask = np.ones((bg_image.shape[0], bg_image.shape[1]), dtype=np.bool_) + save_path = os.path.join(calib_dir, "background_data.npz") + np.savez(save_path, bgrxys=bgrxys, gxyangles=gxyangles, mask=mask) + + +if __name__ == "__main__": + prepare_data() diff --git a/calibration/test_model.py b/calibration/test_model.py new file mode 100644 index 0000000..06659f6 --- /dev/null +++ b/calibration/test_model.py @@ -0,0 +1,97 @@ +import argparse +import os + +import cv2 +import numpy as np +import yaml + +from gs_sdk.gs_device import Camera +from gs_sdk.gs_reconstruct import Reconstructor + +""" +This script tests the calibrated model real-time reconstructing local patches with the sensor. + +Prerequisite: + - Collect data and train the calibration model. +Instructions: + - Connect the sensor to the computer. + - Run this script, wait a bit for background collection, press any key to quit the streaming session. + +Usage: + python test.py --calib_dir CALIB_DIR [--config_path CONFIG_PATH] + +Arguments: + --calib_dir: Path to the directory where the collected data is stored. + --config_path: (Optional) Path to the configuration file about the sensor dimensions. + If not provided, GelSight Mini is assumed. +""" + +config_dir = os.path.join(os.path.dirname(__file__), "../examples/configs") + + +def test_model(): + # Argument Parsers + parser = argparse.ArgumentParser( + description="Read image from the device and reconstruct based on the calibrated model." + ) + parser.add_argument( + "-b", + "--calib_dir", + type=str, + help="place where the calibration data is stored", + ) + parser.add_argument( + "-c", + "--config_path", + type=str, + help="path of the sensor information", + default=os.path.join(config_dir, "gsmini.yaml"), + ) + args = parser.parse_args() + + # Load the device configuration + with open(args.config_path, "r") as f: + config = yaml.safe_load(f) + device_name = config["device_name"] + imgh = config["imgh"] + imgw = config["imgw"] + ppmm = config["ppmm"] + + # Create device and the reconstructor + device = Camera(device_name, imgh, imgw) + device.connect() + model_path = os.path.join(args.calib_dir, "model", "nnmodel.pth") + recon = Reconstructor(model_path, device="cpu") + + # Collect background images + print("Collecting 10 background images, please wait ...") + bg_images = [] + for _ in range(10): + image = device.get_image() + bg_images.append(image) + bg_image = np.mean(bg_images, axis=0).astype(np.uint8) + recon.load_bg(bg_image) + + # Real-time reconstruct + print("\nPrss any key to quit.\n") + while True: + image = device.get_image() + G, H, C = recon.get_surface_info(image, ppmm) + # Create the image for gradient visualization + red = G[:, :, 0] * 255 / 3.0 + 127 + red = np.clip(red, 0, 255) + blue = G[:, :, 1] * 255 / 3.0 + 127 + blue = np.clip(blue, 0, 255) + grad_image = np.stack((blue, np.zeros_like(blue), red), axis=-1).astype(np.uint8) + # Display + cv2.imshow(device_name, grad_image) + key = cv2.waitKey(1) + if key != -1: + break + + device.release() + cv2.destroyAllWindows() + + +if __name__ == "__main__": + test_model() diff --git a/calibration/train_model.py b/calibration/train_model.py new file mode 100644 index 0000000..b8959c5 --- /dev/null +++ b/calibration/train_model.py @@ -0,0 +1,195 @@ +import argparse +import json +import os + +import matplotlib.pyplot as plt +import numpy as np +import torch +from torch import nn +import torch.optim as optim +from torch.utils.data import DataLoader + +from calibration.utils import load_csv_as_dict, transfer_weights +from calibration.models import BGRXYDataset, BGRXYMLPNet_ +from gs_sdk.gs_reconstruct import BGRXYMLPNet + +""" +This script trains the gradient prediction network. +The network is trained as MLP taking the pixel BGRXY as input and predict the gradients gx, gy. + +Prerequisite: + - Tactile images collected using ball indenters with known diameters are collected. + - Collected tactile images are labeled. + - Labeled data are prepared into dataset. + +Usage: + python train.py --calib_dir CALIB_DIR [--n_epochs N_EPOCHS] [--lr LR] [--device {cpu, cuda}] + +Arguments: + --calib_dir: Path to the directory where the collected data will be saved + --n_epochs: (Optional) Number of training epochs. Default is 200. + --lr: (Optional) Learning rate. Default is 0.002. + --device: (Optional) The device to train the network. Can choose between cpu and cuda. Default is cpu. +""" + + +def train_model(): + # Argument Parsers + parser = argparse.ArgumentParser(description="Train the model from BGRXY to gxy.") + parser.add_argument( + "-b", + "--calib_dir", + type=str, + help="place where the calibration data is stored", + ) + parser.add_argument( + "-ne", "--n_epochs", type=int, default=200, help="number of training epochs" + ) + parser.add_argument("-lr", "--lr", type=float, default=0.002, help="learning rate") + parser.add_argument( + "-d", + "--device", + type=str, + choices=["cpu", "cuda"], + default="cpu", + help="the device to train NN", + ) + args = parser.parse_args() + + # Create the model directory + calib_dir = args.calib_dir + model_dir = os.path.join(calib_dir, "model") + if not os.path.isdir(model_dir): + os.makedirs(model_dir) + + # Load the train and test split + data_path = os.path.join(calib_dir, "train_test_split.json") + with open(data_path, "r") as f: + data = json.load(f) + train_reldirs = data["train"] + test_reldirs = data["test"] + + # Load the train and test data including the background data + train_data = {"all_bgrxys": [], "all_gxyangles": []} + for experiment_reldir in train_reldirs: + data_path = os.path.join(calib_dir, experiment_reldir, "data.npz") + if not os.path.isfile(data_path): + raise ValueError("Data file %s does not exist" % data_path) + data = np.load(data_path) + train_data["all_bgrxys"].append(data["bgrxys"][data["mask"]]) + train_data["all_gxyangles"].append(data["gxyangles"][data["mask"]]) + test_data = {"all_bgrxys": [], "all_gxyangles": []} + for experiment_reldir in test_reldirs: + data_path = os.path.join(calib_dir, experiment_reldir, "data.npz") + if not os.path.isfile(data_path): + raise ValueError("Data file %s does not exist" % data_path) + data = np.load(data_path) + test_data["all_bgrxys"].append(data["bgrxys"][data["mask"]]) + test_data["all_gxyangles"].append(data["gxyangles"][data["mask"]]) + #Load background data + bg_path = os.path.join(calib_dir, "background_data.npz") + bg_data = np.load(bg_path) + bgrxys = bg_data["bgrxys"][bg_data["mask"]] + gxyangles = bg_data["gxyangles"][bg_data["mask"]] + perm = np.random.permutation(len(bgrxys)) + n_train = np.sum([len(bgrxys) for bgrxys in train_data["all_bgrxys"]]) // 5 + n_test = np.sum([len(bgrxys) for bgrxys in test_data["all_bgrxys"]]) // 5 + if n_train + n_test > len(bgrxys): + n_train = 4 * len(bgrxys) // 5 + n_test = len(bgrxys) // 5 + train_data["all_bgrxys"].append(bgrxys[perm[:n_train]]) + train_data["all_gxyangles"].append(gxyangles[perm[:n_train]]) + test_data["all_bgrxys"].append(bgrxys[perm[n_train : n_train + n_test]]) + test_data["all_gxyangles"].append(gxyangles[perm[n_train : n_train + n_test]]) + # Construct the train and test dataset + train_bgrxys = np.concatenate(train_data["all_bgrxys"]) + train_gxyangles = np.concatenate(train_data["all_gxyangles"]) + test_bgrxys = np.concatenate(test_data["all_bgrxys"]) + test_gxyangles = np.concatenate(test_data["all_gxyangles"]) + + # Create train and test Dataloader + train_dataset = BGRXYDataset(train_bgrxys, train_gxyangles) + train_dataloader = DataLoader(train_dataset, batch_size=1024, shuffle=True) + test_dataset = BGRXYDataset(test_bgrxys, test_gxyangles) + test_dataloader = DataLoader(test_dataset, batch_size=1024, shuffle=False) + + # Create the MLP Net for training + device = args.device + net = BGRXYMLPNet_().to(device) + criterion = nn.L1Loss() + optimizer = optim.Adam(net.parameters(), lr=args.lr, weight_decay=0.0) + scheduler = optim.lr_scheduler.StepLR(optimizer, step_size=10, gamma=0.5) + + # Create MLP Net in the CNN format for saving + save_net = BGRXYMLPNet().to(device) + + # Initial evaluation + train_mae = evaluate(net, train_dataloader, device) + test_mae = evaluate(net, test_dataloader, device) + naive_mae = np.mean(np.abs(test_gxyangles - np.mean(train_gxyangles, axis=0))) + traj = {"train_maes": [train_mae], "test_maes": [test_mae], "naive_mae": naive_mae} + print("Naive MAE (predict as mean): %.4f" % naive_mae) + print("without train, Train MAE: %.4f, Test MAE: %.4f" % (train_mae, test_mae)) + + # Train the model + for epoch_idx in range(args.n_epochs): + losses = [] + net.train() + for bgrxys, gxyangles in train_dataloader: + bgrxys = bgrxys.to(device) + gxyangles = gxyangles.to(device) + optimizer.zero_grad() + outputs = net(bgrxys) + loss = criterion(outputs, gxyangles) + loss.backward() + optimizer.step() + diffs = outputs - gxyangles + losses.append(np.abs(diffs.cpu().detach().numpy())) + net.eval() + traj["train_maes"].append(np.mean(np.concatenate(losses))) + traj["test_maes"].append(evaluate(net, test_dataloader, device)) + print( + "Epoch %i, Train MAE: %.4f, Test MAE: %.4f" + % (epoch_idx, traj["train_maes"][-1], traj["test_maes"][-1]) + ) + scheduler.step() + + # Save model every 10 steps + if (epoch_idx + 1) % 10 == 0: + # Transfer weights to MLP Net and save + transfer_weights(net, save_net) + save_path = os.path.join(model_dir, "nnmodel.pth") + torch.save(save_net.state_dict(), save_path) + + # Save the training curve + save_path = os.path.join(model_dir, "training_curve.png") + plt.plot(np.arange(len(traj["train_maes"])), traj["train_maes"], color="blue") + plt.plot(np.arange(len(traj["test_maes"])), traj["test_maes"], color="red") + plt.xlabel("Epochs") + plt.ylabel("MAE (rad)") + plt.title("MAE Curve") + plt.savefig(save_path) + plt.close() + + +def evaluate(net, dataloader, device): + """ + Evaluate the network loss on the dataset. + + :param net: nn.Module; the network to evaluate. + :param dataloader: DataLoader; the dataloader for the dataset. + :param device: str; the device to evaluate the network. + """ + losses = [] + for bgrxys, gxyangles in dataloader: + bgrxys = bgrxys.to(device) + gxyangles = gxyangles.to(device) + outputs = net(bgrxys) + diffs = outputs - gxyangles + losses.append(np.abs(diffs.cpu().detach().numpy())) + mae = np.mean(np.concatenate(losses)) + return mae + + +if __name__ == "__main__": + train_model() diff --git a/calibration/utils.py b/calibration/utils.py new file mode 100644 index 0000000..ca0078b --- /dev/null +++ b/calibration/utils.py @@ -0,0 +1,64 @@ +import csv +import numpy as np + + +def load_csv_as_dict(csv_path): + """ + Load the csv file entries as dictionaries. + + :params csv_path: str; the path of the csv file. + :returns: dict; the dictionary of the csv file. + """ + with open(csv_path, "r") as f: + reader = csv.DictReader(f, delimiter=",") + data = list(reader) + keys = reader.fieldnames + data_dict = {} + for key in keys: + data_dict[key] = [] + for line in data: + for key in keys: + data_dict[key].append(line[key]) + return data_dict + + +def transfer_weights(mlp_model, fcn_model): + """ + transfer weights between BGRXYMLPNet_ to BGRXYMLPNet. + + :param mlp_model: BGRXYMLPNet_; the model to transfer from. + :param fcn_model: BGRXYMLPNet; the model to transfer to. + """ + # Copy weights from fc1 to conv1 + fcn_model.conv1.weight.data = mlp_model.fc1.weight.data.view( + fcn_model.conv1.weight.size() + ) + fcn_model.conv1.bias.data = mlp_model.fc1.bias.data + fcn_model.bn1.weight.data = mlp_model.bn1.weight.data + fcn_model.bn1.bias.data = mlp_model.bn1.bias.data + fcn_model.bn1.running_mean = mlp_model.bn1.running_mean + fcn_model.bn1.running_var = mlp_model.bn1.running_var + # Copy weights from fc2 to conv2 + fcn_model.conv2.weight.data = mlp_model.fc2.weight.data.view( + fcn_model.conv2.weight.size() + ) + fcn_model.conv2.bias.data = mlp_model.fc2.bias.data + fcn_model.bn2.weight.data = mlp_model.bn2.weight.data + fcn_model.bn2.bias.data = mlp_model.bn2.bias.data + fcn_model.bn2.running_mean = mlp_model.bn2.running_mean + fcn_model.bn2.running_var = mlp_model.bn2.running_var + # Copy weights from fc3 to conv3 + fcn_model.conv3.weight.data = mlp_model.fc3.weight.data.view( + fcn_model.conv3.weight.size() + ) + fcn_model.conv3.bias.data = mlp_model.fc3.bias.data + fcn_model.bn3.weight.data = mlp_model.bn3.weight.data + fcn_model.bn3.bias.data = mlp_model.bn3.bias.data + fcn_model.bn3.running_mean = mlp_model.bn3.running_mean + fcn_model.bn3.running_var = mlp_model.bn3.running_var + # Copy weights from fc4 to conv4 + fcn_model.conv4.weight.data = mlp_model.fc4.weight.data.view( + fcn_model.conv4.weight.size() + ) + fcn_model.conv4.bias.data = mlp_model.fc4.bias.data + return fcn_model diff --git a/examples/configs/digit.yaml b/examples/configs/digit.yaml new file mode 100644 index 0000000..29f3e88 --- /dev/null +++ b/examples/configs/digit.yaml @@ -0,0 +1,11 @@ +# Device Name +device_name: "DIGIT" +# Pixel Per Millimeter +ppmm: 0.0405 +# Desired Image Width and Height +imgh: 240 +imgw: 320 +# Raw image width, height, and framerate +raw_imgh: 480 +raw_imgw: 640 +framerate: 60 \ No newline at end of file diff --git a/examples/configs/gsmini.yaml b/examples/configs/gsmini.yaml new file mode 100644 index 0000000..0238a21 --- /dev/null +++ b/examples/configs/gsmini.yaml @@ -0,0 +1,11 @@ +# Device Name +device_name: "GelSight Mini" +# Pixel Per Millimeter +ppmm: 0.0634 +# Desired Image Width and Height +imgh: 240 +imgw: 320 +# Raw image width, height, and framerate +raw_imgh: 2464 +raw_imgw: 3280 +framerate: 25 diff --git a/examples/data/background.png b/examples/data/background.png new file mode 100644 index 0000000..68506df Binary files /dev/null and b/examples/data/background.png differ diff --git a/examples/data/bead.png b/examples/data/bead.png new file mode 100644 index 0000000..67510da Binary files /dev/null and b/examples/data/bead.png differ diff --git a/examples/data/key.png b/examples/data/key.png new file mode 100644 index 0000000..4a577ca Binary files /dev/null and b/examples/data/key.png differ diff --git a/examples/data/seed.png b/examples/data/seed.png new file mode 100644 index 0000000..9f3ccb0 Binary files /dev/null and b/examples/data/seed.png differ diff --git a/examples/fast_stream_device.py b/examples/fast_stream_device.py new file mode 100644 index 0000000..80547aa --- /dev/null +++ b/examples/fast_stream_device.py @@ -0,0 +1,50 @@ +import os + +import cv2 +import yaml + +from gs_sdk.gs_device import FastCamera + +""" +This script demonstrates how to use the FastCamera class from the gs_sdk package. + +It loads a configuration file, initializes the FastCamera, and streaming images with low latency. +This script is only for GelSight Mini so far as only GelSight Mini has the frame dropping issue. + +Usage: + python fast_stream_device.py + +Press any key to quit the streaming session. +""" + +config_dir = os.path.join(os.path.dirname(__file__), "configs") + + +def fast_stream_device(): + # Load the device configuration + config_path = os.path.join(config_dir, "gsmini.yaml") + with open(config_path, "r") as f: + config = yaml.safe_load(f) + device_name = config["device_name"] + imgh = config["imgh"] + imgw = config["imgw"] + raw_imgh = config["raw_imgh"] + raw_imgw = config["raw_imgw"] + framerate = config["framerate"] + + # Create device and stream the device + device = FastCamera(device_name, imgh, imgw, raw_imgh, raw_imgw, framerate) + device.connect() + print("\nPrss any key to quit.\n") + while True: + image = device.get_image() + cv2.imshow(device_name, image) + key = cv2.waitKey(1) + if key != -1: + break + device.release() + cv2.destroyAllWindows() + + +if __name__ == "__main__": + fast_stream_device() diff --git a/examples/models/gsmini.pth b/examples/models/gsmini.pth new file mode 100644 index 0000000..422a215 Binary files /dev/null and b/examples/models/gsmini.pth differ diff --git a/examples/reconstruct.py b/examples/reconstruct.py new file mode 100644 index 0000000..6b927b0 --- /dev/null +++ b/examples/reconstruct.py @@ -0,0 +1,78 @@ +import argparse +import os + +import cv2 +import matplotlib.pyplot as plt +import yaml + +from gs_sdk.gs_reconstruct import Reconstructor +from gs_sdk.viz_utils import plot_gradients + +""" +This script demonstrates how to use the Reconstructor class from the gs_sdk package. + +It loads a configuration file, initialize the Reconstructor, reconstruct surface information from images, +and save them to files in the "data/" directory. + +Usage: + python reconstruct.py --device {cuda, cpu} + +Arguments: + --device: The device to load the neural network model. Options are 'cuda' or 'cpu'. +""" + +model_path = os.path.join(os.path.dirname(__file__), "models", "gsmini.pth") +config_path = os.path.join(os.path.dirname(__file__), "configs", "gsmini.yaml") +data_dir = os.path.join(os.path.dirname(__file__), "data") + + +def reconstruct(): + # Argument Parser + parser = argparse.ArgumentParser(description="Reconstruct surface info from data.") + parser.add_argument( + "-d", + "--device", + type=str, + choices=["cuda", "cpu"], + default="cpu", + help="The device to load and run the neural network model.", + ) + args = parser.parse_args() + + # Load the device configuration + with open(config_path, "r") as f: + config = yaml.safe_load(f) + ppmm = config["ppmm"] + + # Create reconstructor + recon = Reconstructor(model_path, device=args.device) + bg_image = cv2.imread(os.path.join(data_dir, "background.png")) + recon.load_bg(bg_image) + + # Reconstruct the surface information from data and save them to files + filenames = ["bead.png", "key.png", "seed.png"] + for filename in filenames: + image = cv2.imread(os.path.join(data_dir, filename)) + G, H, C = recon.get_surface_info(image, ppmm) + + # Plot the surface information + fig, axes = plt.subplots(nrows=2, ncols=2, figsize=(15, 10)) + axes[0, 0].imshow(cv2.cvtColor(image, cv2.COLOR_BGR2RGB)) + axes[0, 0].set_title("GelSight Image") + plot_gradients(fig, axes[0, 1], G[:, :, 0], G[:, :, 1], mask=C, mode="rgb") + axes[0, 1].set_title("Reconstructed Gradients") + axes[1, 0].imshow(H, cmap="jet") + axes[1, 0].set_title("Reconstructed Heights") + axes[1, 1].imshow(C) + axes[1, 1].set_title("Predicted Contact Mask") + for ax in axes.flatten(): + ax.set_xticks([]) + ax.set_yticks([]) + save_path = os.path.join(data_dir, "reconstructed_" + filename) + plt.savefig(save_path) + plt.close() + print("Save results to %s" % save_path) + + +if __name__ == "__main__": + reconstruct() diff --git a/examples/stream_device.py b/examples/stream_device.py new file mode 100644 index 0000000..d3b64a4 --- /dev/null +++ b/examples/stream_device.py @@ -0,0 +1,69 @@ +import argparse +import os + +import cv2 +import yaml + +from gs_sdk.gs_device import Camera + +""" +This script demonstrates how to use the Camera class from the gs_sdk package. + +It loads a configuration file, initializes the Camera, and streaming images. + +Usage: + python stream_device.py --device_name {gsmini, digit} + +Arguments: + --device_name: The name of the device to stream from. Options are 'gsmini' or 'digit'. + Default is 'gsmini'. + +Press any key to quit the streaming session. +""" + +config_dir = os.path.join(os.path.dirname(__file__), "configs") + + +def stream_device(): + # Argument parser + parser = argparse.ArgumentParser(description="Read and show image from the device.") + parser.add_argument( + "-n", + "--device_name", + type=str, + choices=["gsmini", "digit"], + default="gsmini", + help="The name of the device", + ) + args = parser.parse_args() + + # Load the device configuration + if args.device_name == "gsmini": + config_file = "gsmini.yaml" + elif args.device_name == "digit": + config_file = "digit.yaml" + else: + raise ValueError("Unknown device name %s." % args.device_name) + config_path = os.path.join(config_dir, config_file) + with open(config_path, "r") as f: + config = yaml.safe_load(f) + device_name = config["device_name"] + imgh = config["imgh"] + imgw = config["imgw"] + + # Create device and stream the device + device = Camera(device_name, imgh, imgw) + device.connect() + print("\nPrss any key to quit.\n") + while True: + image = device.get_image() + cv2.imshow(device_name, image) + key = cv2.waitKey(1) + if key != -1: + break + device.release() + cv2.destroyAllWindows() + + +if __name__ == "__main__": + stream_device() diff --git a/gs_sdk/__init__.py b/gs_sdk/__init__.py new file mode 100644 index 0000000..e69de29 diff --git a/gs_sdk/gs_device.py b/gs_sdk/gs_device.py new file mode 100644 index 0000000..6a64152 --- /dev/null +++ b/gs_sdk/gs_device.py @@ -0,0 +1,206 @@ +import os +import re +import subprocess + +import cv2 +import ffmpeg +import numpy as np + + +class Camera: + """ + The GelSight Camera Class. + + This class handles camera initialization, image acquisition, and camera release. + Some sensors (GelSight Mini) might experience frame dropping issues, use FastCamera class instead. + """ + + def __init__(self, dev_type, imgh, imgw): + """ + Initialize the camera. + + :param dev_type: str; The type of the camera. + :param imgh: int; The height of the image. + :param imgw: int; The width of the image. + """ + self.dev_type = dev_type + self.dev_id = get_camera_id(self.dev_type) + self.imgh = imgh + self.imgw = imgw + self.cam = None + self.data = None + + def connect(self): + """ + Connect to the camera using cv2 streamer. + """ + self.cam = cv2.VideoCapture(self.dev_id) + self.cam.set(cv2.CAP_PROP_BUFFERSIZE, 1) + if self.cam is None or not self.cam.isOpened(): + print("Warning: unable to open video source %d" % (self.dev_id)) + else: + print("Connect to %s at video source %d" % (self.dev_type, self.dev_id)) + + def get_image(self, flush=False): + """ + Get the image from the camera. + + :param flush: bool; Whether to flush the first few frames. + :return: np.ndarray; The image from the camera. + """ + if flush: + # flush out fist few frames to remove black frames + for i in range(10): + ret, f0 = self.cam.read() + ret, f0 = self.cam.read() + if ret: + f0 = resize_crop(f0, self.imgw, self.imgh) + self.data = f0 + else: + print("ERROR! reading image from video source %d" % (self.dev_id)) + return self.data + + def release(self): + """ + Release the camera resource. + """ + if self.cam is not None: + self.cam.release() + print("Video source %d released." % (self.dev_id)) + else: + print("No camera to release.") + + +class FastCamera: + """ + The GelSight Camera Class with low latency. + + This class handles camera initialization, image acquisition, and camera release with low latency. + """ + + def __init__(self, dev_type, imgh, imgw, raw_imgh, raw_imgw, framerate): + """ + Initialize the low latency camera. Raw camera parameters are required to stream with low latency. + + :param dev_type: str; The type of the camera. + :param imgh: int; The desired height of the image. + :param imgw: int; The desired width of the image. + :param raw_imgh: int; The raw height of the image. + :param raw_imgw: int; The raw width of the image. + :param framerate: int; The frame rate of the camera. + """ + # Raw image size + self.raw_imgh = raw_imgh + self.raw_imgw = raw_imgw + self.raw_size = self.raw_imgh * self.raw_imgw * 3 + self.framerate = framerate + # desired image size + self.imgh = imgh + self.imgw = imgw + # Get camera ID + self.dev_type = dev_type + self.dev_id = get_camera_id(self.dev_type) + self.device = "/dev/video" + str(self.dev_id) + + def connect(self): + """ + Connect to the camera using FFMpeg streamer. + """ + # Command to capture video using ffmpeg and high resolution + self.ffmpeg_command = ( + ffmpeg.input( + self.device, + format="v4l2", + framerate=self.framerate, + video_size="%dx%d" % (self.raw_imgw, self.raw_imgh), + ) + .output("pipe:", format="rawvideo", pix_fmt="bgr24") + .global_args("-fflags", "nobuffer") + .global_args("-flags", "low_delay") + .global_args("-fflags", "+genpts") + .global_args("-rtbufsize", "0") + .compile() + ) + self.process = subprocess.Popen( + self.ffmpeg_command, stdout=subprocess.PIPE, stderr=subprocess.PIPE + ) + # Warm-up phase: discard the first few frames + print("Warming up the camera...") + warm_up_frames = 100 + for _ in range(warm_up_frames): + self.process.stdout.read(self.raw_size) + print("Camera ready for use!") + + def get_image(self): + """ + Get the image from the camera from raw data stream. + + :return: np.ndarray; The image from the camera. + """ + raw_frame = self.process.stdout.read(self.raw_size) + frame = np.frombuffer(raw_frame, np.uint8).reshape( + (self.raw_imgh, self.raw_imgw, 3) + ) + frame = resize_crop(frame, self.imgw, self.imgh) + return frame + + def release(self): + """ + Release the camera resource. + """ + self.process.stdout.close() + self.process.wait() + + +def get_camera_id(camera_name): + """ + Find the camera ID that has the corresponding camera name. + + :param camera_name: str; The name of the camera. + :return: int; The camera ID. + """ + cam_num = None + for file in os.listdir("/sys/class/video4linux"): + real_file = os.path.realpath("/sys/class/video4linux/" + file + "/name") + with open(real_file, "rt") as name_file: + name = name_file.read().rstrip() + if camera_name in name: + cam_num = int(re.search("\d+$", file).group(0)) + found = "FOUND!" + else: + found = " " + print("{} {} -> {}".format(found, file, name)) + + return cam_num + + +def resize_crop(img, imgw, imgh): + """ + Resize and crop the image to the desired size. + + :param img: np.ndarray; The image to resize and crop. + :param imgw: int; The width of the desired image. + :param imgh: int; The height of the desired image. + :return: np.ndarray; The resized and cropped image. + """ + # remove 1/7th of border from each size + border_size_x, border_size_y = int(img.shape[0] * (1 / 7)), int( + np.floor(img.shape[1] * (1 / 7)) + ) + cropped_imgh = img.shape[0] - 2 * border_size_x + cropped_imgw = img.shape[1] - 2 * border_size_y + # Extra cropping to maintain aspect ratio + extra_border_h = 0 + extra_border_w = 0 + if cropped_imgh * imgw / imgh > cropped_imgw + 1e-8: + extra_border_h = int(cropped_imgh - cropped_imgw * imgh / imgw) + elif cropped_imgh * imgw / imgh < cropped_imgw - 1e-8: + extra_border_w = int(cropped_imgw - cropped_imgh * imgw / imgh) + # keep the ratio the same as the original image size + img = img[ + border_size_x + extra_border_h : img.shape[0] - border_size_x, + border_size_y + extra_border_w : img.shape[1] - border_size_y, + ] + # final resize for the desired image size + img = cv2.resize(img, (imgw, imgh)) + return img diff --git a/gs_sdk/gs_reconstruct.py b/gs_sdk/gs_reconstruct.py new file mode 100644 index 0000000..3f91f03 --- /dev/null +++ b/gs_sdk/gs_reconstruct.py @@ -0,0 +1,214 @@ +import math +import os + +import cv2 +import numpy as np +from scipy import fftpack +import torch +import torch.nn as nn +import torch.nn.functional as F + + +class BGRXYMLPNet(nn.Module): + """ + The Neural Network architecture for GelSight calibration. + + This class uses 1-by-1 convolution, which is technically the same as using MLP. + """ + + def __init__(self): + super(BGRXYMLPNet, self).__init__() + input_channels = 5 + self.conv1 = nn.Conv2d(input_channels, 128, kernel_size=1) + self.bn1 = nn.BatchNorm2d(128) + self.conv2 = nn.Conv2d(128, 32, kernel_size=1) + self.bn2 = nn.BatchNorm2d(32) + self.conv3 = nn.Conv2d(32, 32, kernel_size=1) + self.bn3 = nn.BatchNorm2d(32) + self.conv4 = nn.Conv2d(32, 2, kernel_size=1) + + def forward(self, x): + x = F.relu(self.bn1(self.conv1(x))) + x = F.relu(self.bn2(self.conv2(x))) + x = F.relu(self.bn3(self.conv3(x))) + x = self.conv4(x) + return x + + +class Reconstructor: + """ + The GelSight reconstruction class. + + This class handles 3D reconstruction from calibrated GelSight images. + """ + + def __init__(self, model_path, contact_mode="standard", device="cpu"): + """ + Initialize the reconstruction model. + Contact mode "flat" means the object in contact is flat, so a different threshold + is used to determine contact mask. + + :param model_path: str; the path of the calibrated neural network model. + :param contact_mode: str {"standard", "flat"}; the mode to get the contact mask. + :param device: str {"cuda", "cpu"}; the device to run the model. + """ + self.model_path = model_path + self.contact_mode = contact_mode + self.device = device + self.bg_image = None + # Load the gxy model + if not os.path.isfile(model_path): + raise ValueError("Error opening %s, file does not exist" % model_path) + self.gxy_net = BGRXYMLPNet() + self.gxy_net.load_state_dict(torch.load(model_path), self.device) + self.gxy_net.eval() + + def load_bg(self, bg_image): + """ + Load the background image. + + :param bg_image: np.array (H, W, 3); the background image. + """ + self.bg_image = bg_image + + # Calculate the gradients of the background + bgrxys = image2bgrxys(bg_image) + bgrxys = bgrxys.transpose(2, 0, 1) + features = torch.from_numpy(bgrxys[np.newaxis, :, :, :]).float().to(self.device) + with torch.no_grad(): + gxyangles = self.gxy_net(features) + gxyangles = gxyangles[0].cpu().detach().numpy() + self.bg_G = np.tan(gxyangles.transpose(1, 2, 0)) + + def get_surface_info(self, image, ppmm): + """ + Get the surface information including gradients (G), height map (H), and contact mask (C). + + :param image: np.array (H, W, 3); the gelsight image. + :param ppmm: float; the pixel per mm. + :return G: np.array (H, W, 2); the gradients. + H: np.array (H, W); the height map. + C: np.array (H, W); the contact mask. + """ + # Calculate the gradients + bgrxys = image2bgrxys(image) + bgrxys = bgrxys.transpose(2, 0, 1) + features = torch.from_numpy(bgrxys[np.newaxis, :, :, :]).float().to(self.device) + with torch.no_grad(): + gxyangles = self.gxy_net(features) + gxyangles = gxyangles[0].cpu().detach().numpy() + G = np.tan(gxyangles.transpose(1, 2, 0)) + if self.bg_image is not None: + G = G - self.bg_G + else: + raise ValueError("Background image is not loaded.") + + # Calculate the height map + H = poisson_dct_neumaan(G[:, :, 0], G[:, :, 1]).astype(np.float32) + + # Calculate the contact mask + if self.contact_mode == "standard": + # Find the contact mask based on color difference + diff_image = image.astype(np.float32) - self.bg_image.astype(np.float32) + color_mask = np.linalg.norm(diff_image, axis=-1) > 15 + color_mask = cv2.dilate( + color_mask.astype(np.uint8), np.ones((7, 7), np.uint8) + ) + color_mask = cv2.erode( + color_mask.astype(np.uint8), np.ones((15, 15), np.uint8) + ) + + # Filter by height + cutoff = np.percentile(H, 85) - 0.2 / ppmm + height_mask = H < cutoff + C = np.logical_and(color_mask, height_mask) + elif self.contact_mode == "flat": + # Find the contact mask based on color difference + diff_image = image.astype(np.float32) - self.bg_image.astype(np.float32) + color_mask = np.linalg.norm(diff_image, axis=-1) > 10 + color_mask = cv2.dilate( + color_mask.astype(np.uint8), np.ones((15, 15), np.uint8) + ) + C = cv2.erode( + color_mask.astype(np.uint8), np.ones((25, 25), np.uint8) + ).astype(np.bool_) + + return G, H, C + + +def image2bgrxys(image): + """ + Convert a bgr image to bgrxy feature. + + :param image: np.array (H, W, 3); the bgr image. + :return: np.array (H, W, 5); the bgrxy feature. + """ + xys = np.dstack( + np.meshgrid(np.arange(image.shape[1]), np.arange(image.shape[0]), indexing="xy") + ) + xys = xys.astype(np.float32) / np.array([image.shape[1], image.shape[0]]) + bgrs = image.copy() / 255 + bgrxys = np.concatenate([bgrs, xys], axis=2) + return bgrxys + + +def poisson_dct_neumaan(gx, gy): + """ + 2D integration of depth from gx, gy using Poisson solver. + + :param gx: np.array (H, W); the x gradient. + :param gy: np.array (H, W); the y gradient. + :return: np.array (H, W); the depth map. + """ + # Compute Laplacian + gxx = 1 * ( + gx[:, (list(range(1, gx.shape[1])) + [gx.shape[1] - 1])] + - gx[:, ([0] + list(range(gx.shape[1] - 1)))] + ) + gyy = 1 * ( + gy[(list(range(1, gx.shape[0])) + [gx.shape[0] - 1]), :] + - gy[([0] + list(range(gx.shape[0] - 1))), :] + ) + f = gxx + gyy + + # Right hand side of the boundary condition + b = np.zeros(gx.shape) + b[0, 1:-2] = -gy[0, 1:-2] + b[-1, 1:-2] = gy[-1, 1:-2] + b[1:-2, 0] = -gx[1:-2, 0] + b[1:-2, -1] = gx[1:-2, -1] + b[0, 0] = (1 / np.sqrt(2)) * (-gy[0, 0] - gx[0, 0]) + b[0, -1] = (1 / np.sqrt(2)) * (-gy[0, -1] + gx[0, -1]) + b[-1, -1] = (1 / np.sqrt(2)) * (gy[-1, -1] + gx[-1, -1]) + b[-1, 0] = (1 / np.sqrt(2)) * (gy[-1, 0] - gx[-1, 0]) + + # Modification near the boundaries to enforce the non-homogeneous Neumann BC (Eq. 53 in [1]) + f[0, 1:-2] = f[0, 1:-2] - b[0, 1:-2] + f[-1, 1:-2] = f[-1, 1:-2] - b[-1, 1:-2] + f[1:-2, 0] = f[1:-2, 0] - b[1:-2, 0] + f[1:-2, -1] = f[1:-2, -1] - b[1:-2, -1] + + # Modification near the corners (Eq. 54 in [1]) + f[0, -1] = f[0, -1] - np.sqrt(2) * b[0, -1] + f[-1, -1] = f[-1, -1] - np.sqrt(2) * b[-1, -1] + f[-1, 0] = f[-1, 0] - np.sqrt(2) * b[-1, 0] + f[0, 0] = f[0, 0] - np.sqrt(2) * b[0, 0] + + # Cosine transform of f + tt = fftpack.dct(f, norm="ortho") + fcos = fftpack.dct(tt.T, norm="ortho").T + + # Cosine transform of z (Eq. 55 in [1]) + (x, y) = np.meshgrid(range(1, f.shape[1] + 1), range(1, f.shape[0] + 1), copy=True) + denom = 4 * ( + (np.sin(0.5 * math.pi * x / (f.shape[1]))) ** 2 + + (np.sin(0.5 * math.pi * y / (f.shape[0]))) ** 2 + ) + + # Inverse Discrete cosine Transform + f = -fcos / denom + tt = fftpack.idct(f, norm="ortho") + img_tt = fftpack.idct(tt.T, norm="ortho").T + img_tt = img_tt.mean() + img_tt + + return img_tt diff --git a/gs_sdk/viz_utils.py b/gs_sdk/viz_utils.py new file mode 100644 index 0000000..95d04de --- /dev/null +++ b/gs_sdk/viz_utils.py @@ -0,0 +1,44 @@ +import matplotlib.pyplot as plt +import numpy as np + + +def plot_gradients(fig, ax, gx, gy, mask=None, mode="rgb", **kwargs): + """ + Plot the gradients. + + :params fig: plt.figure; the figure to plot the gradients. + :params ax: plt.axis; the axis to plot the gradients. + :params gx: np.array (H, W); the x gradient. + :params gy: np.array (H, W); the y gradient. + :params mask: np.array (H, W); the mask for gradients to be plotted + :params mode: str {"rgb", "quiver"}; the mode to plot the gradients. + """ + if mode == "rgb": + # Plot the gradient in red and blue + grad_range = kwargs.get("grad_range", 3.0) + red = gx * 255 / grad_range + 127 + red = np.clip(red, 0, 255) + blue = gy * 255 / grad_range + 127 + blue = np.clip(blue, 0, 255) + image = np.stack((red, np.zeros_like(red), blue), axis=-1).astype(np.uint8) + if mask is not None: + image[np.logical_not(mask)] = np.array([127, 0, 127]) + ax.imshow(image) + elif mode == "quiver": + # Plot the gradient in quiver + n_skip = kwargs.get("n_skip", 5) + quiver_scale = kwargs.get("quiver_scale", 10.0) + imgh, imgw = gx.shape + X, Y = np.meshgrid(np.arange(imgw)[::n_skip], np.arange(imgh)[::n_skip]) + U = gx[::n_skip, ::n_skip] * quiver_scale + V = -gy[::n_skip, ::n_skip] * quiver_scale + if mask is None: + mask = np.ones_like(gx) + else: + mask = np.copy(mask) + mask = mask[::n_skip, ::n_skip] + ax.quiver(X[mask], Y[mask], U[mask], V[mask], units="xy", scale=1, color="red") + ax.set_xlim(0, imgw) + ax.set_ylim(imgh, 0) + else: + raise ValueError("Unknown plot gradient mode %s" % mode) diff --git a/setup.py b/setup.py new file mode 100644 index 0000000..2fc1fc0 --- /dev/null +++ b/setup.py @@ -0,0 +1,34 @@ +from setuptools import setup, find_packages + +setup( + name="gs_sdk", + version="0.1.0", + description="SDK for GelSight sensors usage, reconstruction, and calibration.", + author="Hung-Jui Huang, Ruihan Gao", + author_email="hungjuih@andrew.cmu.edu, ruihang@andrew.cmu.edu", + packages=find_packages(), + install_requires=[ + "pillow==10.0.0", + "numpy==1.26.4", + "opencv-python>=4.9.0", + "scipy>=1.13.1", + "torch>=2.1.0", + "PyYaml>=6.0.1", + "matplotlib>=3.9.0", + "ffmpeg-python", + "nanogui" + ], + python_requires=">=3.9", + entry_points={ + 'console_scripts': [ + 'collect_data=calibration.collect_data:collect_data', + 'label_data=calibration.label_data:label_data', + 'prepare_data=calibration.prepare_data:prepare_data', + 'train_model=calibration.train_model:train_model', + 'test_model=calibration.test_model:test_model', + ], + }, + classifiers=[ + "Programming Language :: Python :: 3", + ], +)