This software is designed to quickly generate software that can generate datasets for MediaPipe ModelMaker. Currently, the software provides two formats: manually generated and automatically generated datasets.
- The user should first ensure that the machine is connected to a number of available cameras. The software will use the VedioCapture library of Opencv-python to open your target camera for reading.
- Please enter the Label tag of the category to be generated this time. Mediapipe only needs a folder expressing the label to load the target hand image set.
- Select the base folder of the folder. All categories of datasets will be generated here. Take the author's test as an example:
tree .
- BaseDir <- select the existing dir: "BaseDir" as the base dirent
|--open_palm # Auto generate
|--closed_fist # Auto generate
- Adjust the mode you want to use! Automatic or manual; For automatic mode, please adjust the period of a single capture. This software supports 1s - 10s
- Open the target camera. The manual mode requires a click, and the data set will be captured when the click occurs. This method is suitable for generating a more sophisticated data set. Meanwhile, the automatic mode will trigger the capture of the data set once according to the period. Of course, you can also click manual collection to insert the data set yourself, instead of waiting for the set number of seconds
- Turn off the camera and go to the specified folder to get the data set
Please refer to requirements.txt generated by pipreq. Only three main dependencies are used to make the software
pip install pyqt6 # Main Ui framework dependencies
pip install mediapipe # Automatic capture hand's landmarks
pip install pygrabber # Get the list of available cameras on the machine
All instructions required are listes in build.md, see the markdown file for detail if you havc self-defined pack up requists!