A classification model is trained, then using a GRADCAM method is employed to extract masks used to train a segementation model. The Pinot Noir Grapes dataset with a yolov8 format is preprocessed to be used as example.
To preprocess the Pinot Noir Grapes dataset from the yolov8 format:
python ./tools/preprocessing.py --dataset_path --processed_dataset_path --img_size
with processed_dataset_path as the output processed dataset path.
To train the binary classifier model:
python ./classification_model/train.py --dataset_root_path
with dataset_root_path the processed dataset path.
To extract the masks from the classifier:
python ./mask_extractor/generate_mask.py --dataset_path --segmentation_dataset_path --model_weights
with segmentation_dataset_path the path to the output dataset with masks and model_weights the path to the classifier trained.
To traing the segmentation model:
python ./segmentation_model/train.py --dataset_path --run_files
with dataset_path the path to the dataset with masks and run_files the path to save runs files.
Examples from the original images, the mask generated with the GRADCAM and the segmentation model is shown below.
