- requests - used in image download
- pillow - image preprocessing, conversion
- libwebp - library for reading .webp
- numpy
- tensorflow (tensorflow-gpu) - backend for keras
- keras - API that is used to construct model
The downloadImg.py demonstrates how we automatically download, categorize and convert images using links in dfClean.csv, which contains the sources of training images. These images will be put into their corresponding categories automatically. Note that dfClean.csv includes images from sources that are protected by NDA, thus is not shared on the repo.
Training.ipynb includes the training code for image recognition models. It demonstrates the process of training ResNet50 and VGG16 models on both binary and multi-class classification.
Demo.ipynb includes the demo code for testing the trained model on given images. It loads models previously trained in Training.ipynb and test them on chosen image (stored in \img). Here we include a VGG16 model (best_vgg.hdf5) that is trained on Similar data. Here we also use labelDecode.py as helper function to decode the prediction array.