An object detection and segmentation for Oysters. It uses Detecron2 implementation.
Modify oysterd/config.py to modify the training and evaluation data folder, the folder containing the video to be used for inference, and the network structure of the model defined in Detectron2 Model Zoo. You can assign multiple network structures here.
- Config.folders['data']: training and evaluation folder
- Config.folders['infer']: inference folder
- Config.folders['output']: output folder
- Config.folders['weights']: folder where the trained network stored
- Config.config_file: a list of models defined in Detectron2 Model Zoo. You can select which model you want to train by passing its number in this list as the
<model_number>argument topython oysterd/train.py -i <model_number> - Config.resume: whether or not to resume/restart the training
Either run python oysterd/train.py -i <model_number> to train a single model <model_number> defined in ysterd/config.py, or run sh run_trains.sh for training multiple models.
The -p argument of oysterd/train.py determines whether or not predict the images in the evalution folder after training.
Use python oysterd/infer.py -i <model_number> if you defined the video path in the oysterd/config.py. You can also pass the folder path by -f argument.
Use python oysterd/infer_video.py -i <model_number> if you defined the video path in the oysterd/config.py. You can also pass the video path by -p argument.
