Instance Segmentation

Start detectron2 training from /AnomaliesRecycling/

python image_segmentation/train_net.py

Models are saved into <Constants.STORE_DIR>/segmentation_experiments/.

Adjust the network configuration in /configs/.

Change the optimizer and evaluator in PolysecureTrainer.py

Default model: Mask R-CNN with FPN and ResNet50 backbone. SGD optimizer. Annealing LR=0.001 in steps (3000, 4000). For more info, see detectron2