Add PNG reading support

.circleci/unittest/linux/scripts/environment.yml

@@ -6,6 +6,7 @@ dependencies:
   - pytest-cov
   - codecov
   - pip
+  - libpng
   - ca-certificates
   - pip:
     - future

.circleci/unittest/windows/scripts/environment.yml

@@ -6,9 +6,10 @@ dependencies:
   - pytest-cov
   - codecov
   - pip
+  - libpng
   - ca-certificates
   - pip:
     - future
     - pillow>=4.1.1
     - scipy==1.4.1
-    - av
+    - av

.gitignore

@@ -21,3 +21,5 @@ htmlcov
 *.swo
 gen.yml
 .mypy_cache
+.vscode/
+*.orig

.travis.yml

@@ -13,6 +13,7 @@ jobs:
 
 before_install:
   - sudo apt-get update
+  - sudo apt-get install -y libpng-dev
   - wget https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh -O miniconda.sh;
   - bash miniconda.sh -b -p $HOME/miniconda
   - export PATH="$HOME/miniconda/bin:$PATH"

CMakeLists.txt

@@ -11,22 +11,29 @@ if(WITH_CUDA)
 endif()
 
 find_package(Python3 COMPONENTS Development)
+
 find_package(Torch REQUIRED)
+find_package(PNG REQUIRED)
+
 
 file(GLOB HEADERS torchvision/csrc/*.h)
-file(GLOB OPERATOR_SOURCES torchvision/csrc/cpu/*.h torchvision/csrc/cpu/*.cpp torchvision/csrc/*.cpp)
+# Image extension
+file(GLOB IMAGE_HEADERS torchvision/csrc/cpu/image/*.h)
+file(GLOB IMAGE_SOURCES torchvision/csrc/cpu/image/*.cpp)
+file(GLOB OPERATOR_SOURCES torchvision/csrc/cpu/*.h torchvision/csrc/cpu/*.cpp ${IMAGE_HEADERS} ${IMAGE_SOURCES} ${HEADERS} torchvision/csrc/*.cpp)
 if(WITH_CUDA)
   file(GLOB OPERATOR_SOURCES ${OPERATOR_SOURCES} torchvision/csrc/cuda/*.h torchvision/csrc/cuda/*.cu)
 endif()
 file(GLOB MODELS_HEADERS torchvision/csrc/models/*.h)
 file(GLOB MODELS_SOURCES torchvision/csrc/models/*.h torchvision/csrc/models/*.cpp)
 
-add_library(${PROJECT_NAME} SHARED ${MODELS_SOURCES} ${OPERATOR_SOURCES})
-target_link_libraries(${PROJECT_NAME} PRIVATE ${TORCH_LIBRARIES} Python3::Python)
+add_library(${PROJECT_NAME} SHARED ${MODELS_SOURCES} ${OPERATOR_SOURCES} ${IMAGE_SOURCES})
+target_link_libraries(${PROJECT_NAME} PRIVATE ${TORCH_LIBRARIES} ${PNG_LIBRARY} Python3::Python)
+# target_link_libraries(${PROJECT_NAME} PRIVATE ${PNG_LIBRARY} Python3::Python)
 set_target_properties(${PROJECT_NAME} PROPERTIES EXPORT_NAME TorchVision)
 
 target_include_directories(${PROJECT_NAME} INTERFACE
-  $<BUILD_INTERFACE:${HEADERS}>
+  $<BUILD_INTERFACE:${HEADERS}:${PNG_INCLUDE_DIR}>
   $<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}>)
 
 include(GNUInstallDirs)
@@ -61,7 +68,7 @@ install(FILES
   DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/${PROJECT_NAME}/cpu)
 if(WITH_CUDA)
   install(FILES
-    torchvision/csrc/cuda/vision_cuda.h 
+    torchvision/csrc/cuda/vision_cuda.h
     DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/${PROJECT_NAME}/cuda)
 endif()
 install(FILES ${MODELS_HEADERS} DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/${PROJECT_NAME}/models)

README.rst

@@ -78,13 +78,19 @@ Torchvision currently supports the following image backends:
 
 * `accimage`_ - if installed can be activated by calling :code:`torchvision.set_image_backend('accimage')`
 
+* `libpng`_ - can be installed via conda :code:`conda install libpng` or any of the package managers for debian-based and RHEL-based Linux distributions.
+
+**Notes:** ``libpng`` must be available at compilation time in order to be available. Make sure that it is available on the standard library locations,
+otherwise, add the include and library paths in the environment variables ``TORCHVISION_INCLUDE`` and ``TORCHVISION_LIBRARY``, respectively.
+
+.. _libpng : http://www.libpng.org/pub/png/libpng.html
 .. _Pillow : https://python-pillow.org/
 .. _Pillow-SIMD : https://github.com/uploadcare/pillow-simd
 .. _accimage: https://github.com/pytorch/accimage
 
 C++ API
 =======
-TorchVision also offers a C++ API that contains C++ equivalent of python models. 
+TorchVision also offers a C++ API that contains C++ equivalent of python models.
 
 Installation From source:
 
@@ -94,7 +100,7 @@ Installation From source:
     cd build
     # Add -DWITH_CUDA=on support for the CUDA if needed
     cmake ..
-    make 
+    make
     make install
 
 Once installed, the library can be accessed in cmake (after properly configuring ``CMAKE_PREFIX_PATH``) via the :code:`TorchVision::TorchVision` target:

packaging/build_wheel.sh

@@ -10,6 +10,23 @@ setup_wheel_python
 pip_install numpy pyyaml future ninja
 setup_pip_pytorch_version
 python setup.py clean
+
+# Copy binaries to be included in the wheel distribution
+if [[ "$(uname)" == Darwin || "$OSTYPE" == "msys" ]]; then
+    python_exec="$(which python)"
+    bin_path=$(dirname $python_exec)
+    env_path=$(dirname $bin_path)
+    if [[ "$(uname)" == Darwin ]]; then
+        # Include LibPNG
+        cp "$env_path/lib/libpng16.dylib" torchvision
+    else
+        cp "$bin_path/Library/bin/libpng16.dll" torchvision
+    fi
+else
+    # Include LibPNG
+    cp "/usr/lib64/libpng.so" torchvision
+fi
+
 if [[ "$OSTYPE" == "msys" ]]; then
     IS_WHEEL=1 "$script_dir/windows/internal/vc_env_helper.bat" python setup.py bdist_wheel
 else

packaging/pkg_helpers.bash

@@ -170,7 +170,11 @@ setup_wheel_python() {
     conda env remove -n "env$PYTHON_VERSION" || true
     conda create -yn "env$PYTHON_VERSION" python="$PYTHON_VERSION"
     conda activate "env$PYTHON_VERSION"
+    # Install libpng from Anaconda (defaults)
+    conda install libpng -y
   else
+    # Install native CentOS libPNG
+    yum install -y libpng-devel
     case "$PYTHON_VERSION" in
       2.7)
         if [[ -n "$UNICODE_ABI" ]]; then

packaging/torchvision/meta.yaml

@@ -8,6 +8,7 @@ source:
 requirements:
   build:
     - {{ compiler('c') }} # [win]
+    - libpng
 
   host:
     - python
@@ -18,6 +19,7 @@ requirements:
 
   run:
     - python
+    - libpng
     - pillow >=4.1.1
     - numpy >=1.11
     {{ environ.get('CONDA_PYTORCH_CONSTRAINT') }}

setup.py

@@ -3,7 +3,7 @@
 import re
 import sys
 from setuptools import setup, find_packages
-from pkg_resources import get_distribution, DistributionNotFound
+from pkg_resources import parse_version, get_distribution, DistributionNotFound
 import subprocess
 import distutils.command.clean
 import distutils.spawn
@@ -76,6 +76,65 @@ def write_version_file():
 requirements.append(pillow_req + pillow_ver)
 
 
+def find_library(name, vision_include):
+    this_dir = os.path.dirname(os.path.abspath(__file__))
+    build_prefix = os.environ.get('BUILD_PREFIX', None)
+    is_conda_build = build_prefix is not None
+
+    library_found = False
+    conda_installed = False
+    lib_folder = None
+    include_folder = None
+    library_header = '{0}.h'.format(name)
+
+    print('Running build on conda-build: {0}'.format(is_conda_build))
+    if is_conda_build:
+        # Add conda headers/libraries
+        if os.name == 'nt':
+            build_prefix = os.path.join(build_prefix, 'Library')
+        include_folder = os.path.join(build_prefix, 'include')
+        lib_folder = os.path.join(build_prefix, 'lib')
+        library_header_path = os.path.join(
+            include_folder, library_header)
+        library_found = os.path.isfile(library_header_path)
+        conda_installed = library_found
+    else:
+        # Check if using Anaconda to produce wheels
+        conda = distutils.spawn.find_executable('conda')
+        is_conda = conda is not None
+        print('Running build on conda: {0}'.format(is_conda))
+        if is_conda:
+            python_executable = sys.executable
+            py_folder = os.path.dirname(python_executable)
+            if os.name == 'nt':
+                env_path = os.path.join(py_folder, 'Library')
+            else:
+                env_path = os.path.dirname(py_folder)
+            lib_folder = os.path.join(env_path, 'lib')
+            include_folder = os.path.join(env_path, 'include')
+            library_header_path = os.path.join(
+                include_folder, library_header)
+            library_found = os.path.isfile(library_header_path)
+            conda_installed = library_found
+
+    if not library_found:
+        if sys.platform == 'linux':
+            library_found = os.path.exists('/usr/include/{0}'.format(
+                library_header))
+            library_found = library_found or os.path.exists(
+                '/usr/local/include/{0}'.format(library_header))
+        else:
+            # Lookup in TORCHVISION_INCLUDE or in the package file
+            package_path = [os.path.join(this_dir, 'torchvision')]
+            for folder in vision_include + package_path:
+                candidate_path = os.path.join(folder, library_header)
+                library_found = os.path.exists(candidate_path)
+                if library_found:
+                    break
+
+    return library_found, conda_installed, include_folder, lib_folder
+
+
 def get_extensions():
     this_dir = os.path.dirname(os.path.abspath(__file__))
     extensions_dir = os.path.join(this_dir, 'torchvision', 'csrc')
@@ -171,6 +230,76 @@ def get_extensions():
             )
         )
 
+    # ------------------- Torchvision extra extensions ------------------------
+    vision_include = os.environ.get('TORCHVISION_INCLUDE', None)
+    vision_library = os.environ.get('TORCHVISION_LIBRARY', None)
+    vision_include = (vision_include.split(os.pathsep)
+                      if vision_include is not None else [])
+    vision_library = (vision_library.split(os.pathsep)
+                      if vision_library is not None else [])
+    include_dirs += vision_include
+    library_dirs = vision_library
+
+    # Image reading extension
+    image_macros = []
+    image_include = [extensions_dir]
+    image_library = []
+    image_link_flags = []
+
+    # Locating libPNG
+    libpng = distutils.spawn.find_executable('libpng-config')
+    pngfix = distutils.spawn.find_executable('pngfix')
+    png_found = libpng is not None or pngfix is not None
+    image_macros += [('PNG_FOUND', str(int(png_found)))]
+    print('PNG found: {0}'.format(png_found))
+    if png_found:
+        if libpng is not None:
+            # Linux / Mac
+            png_version = subprocess.run([libpng, '--version'],
+                                         stdout=subprocess.PIPE)
+            png_version = png_version.stdout.strip().decode('utf-8')
+            print('libpng version: {0}'.format(png_version))
+            png_version = parse_version(png_version)
+            if png_version >= parse_version("1.6.0"):
+                print('Building torchvision with PNG image support')
+                png_lib = subprocess.run([libpng, '--libdir'],
+                                         stdout=subprocess.PIPE)
+                png_include = subprocess.run([libpng, '--I_opts'],
+                                             stdout=subprocess.PIPE)
+                png_include = png_include.stdout.strip().decode('utf-8')
+                _, png_include = png_include.split('-I')
+                print('libpng include path: {0}'.format(png_include))
+                image_library += [png_lib.stdout.strip().decode('utf-8')]
+                image_include += [png_include]
+                image_link_flags.append('png')
+            else:
+                print('libpng installed version is less than 1.6.0, '
+                      'disabling PNG support')
+                png_found = False
+        else:
+            # Windows
+            png_lib = os.path.join(
+                os.path.dirname(os.path.dirname(pngfix)), 'lib')
+            png_include = os.path.join(os.path.dirname(
+                os.path.dirname(pngfix)), 'include', 'libpng16')
+            image_library += [png_lib]
+            image_include += [png_include]
+            image_link_flags.append('libpng')
+
+    image_path = os.path.join(extensions_dir, 'cpu', 'image')
+    image_src = glob.glob(os.path.join(image_path, '*.cpp'))
+
+    if png_found:
+        ext_modules.append(extension(
+            'torchvision.image',
+            image_src,
+            include_dirs=image_include + include_dirs + [image_path],
+            library_dirs=image_library + library_dirs,
+            define_macros=image_macros,
+            libraries=image_link_flags,
+            extra_compile_args=extra_compile_args
+        ))
+
     ffmpeg_exe = distutils.spawn.find_executable('ffmpeg')
     has_ffmpeg = ffmpeg_exe is not None
 
@@ -243,7 +372,9 @@ def run(self):
 
     # Package info
     packages=find_packages(exclude=('test',)),
-
+    package_data={
+        package_name: ['*.dll', '*.dylib', '*.so']
+    },
     zip_safe=False,
     install_requires=requirements,
     extras_require={

test/test_image.py

@@ -0,0 +1,46 @@
+import os
+import unittest
+import sys
+
+import torch
+import torchvision
+from PIL import Image
+from torchvision.io.image import read_png, decode_png
+import numpy as np
+
+IMAGE_ROOT = os.path.join(os.path.dirname(os.path.abspath(__file__)), "assets")
+IMAGE_DIR = os.path.join(IMAGE_ROOT, "fakedata", "imagefolder")
+
+
+def get_images(directory, img_ext):
+    assert os.path.isdir(directory)
+    for root, _, files in os.walk(directory):
+        for fl in files:
+            _, ext = os.path.splitext(fl)
+            if ext == img_ext:
+                yield os.path.join(root, fl)
+
+
+class ImageTester(unittest.TestCase):
+    def test_read_png(self):
+        # Check across .png
+        for img_path in get_images(IMAGE_DIR, ".png"):
+            img_pil = torch.from_numpy(np.array(Image.open(img_path)))
+            img_lpng = read_png(img_path)
+            self.assertTrue(img_lpng.equal(img_pil))
+
+    def test_decode_png(self):
+        for img_path in get_images(IMAGE_DIR, ".png"):
+            img_pil = torch.from_numpy(np.array(Image.open(img_path)))
+            size = os.path.getsize(img_path)
+            img_lpng = decode_png(torch.from_file(img_path, dtype=torch.uint8, size=size))
+            self.assertTrue(img_lpng.equal(img_pil))
+
+            with self.assertRaises(ValueError):
+                decode_png(torch.empty((), dtype=torch.uint8))
+            with self.assertRaises(RuntimeError):
+                decode_png(torch.randint(3, 5, (300,), dtype=torch.uint8))
+
+
+if __name__ == '__main__':
+    unittest.main()

torchvision/csrc/cpu/image/image.cpp

@@ -0,0 +1,16 @@
+
+#include "image.h"
+#include <ATen/ATen.h>
+#include <Python.h>
+
+// If we are in a Windows environment, we need to define
+// initialization functions for the _custom_ops extension
+#ifdef _WIN32
+PyMODINIT_FUNC PyInit_image(void) {
+  // No need to do anything.
+  return NULL;
+}
+#endif
+
+static auto registry =
+    torch::RegisterOperators().op("image::decode_png", &decodePNG);

torchvision/csrc/cpu/image/image.h

@@ -0,0 +1,7 @@
+
+#pragma once
+
+// Comment
+#include <torch/script.h>
+#include <torch/torch.h>
+#include "readpng_cpu.h"