Aerial Imagery with .TIF Extension
Geometry Masks in .geojson Format
SAM Model Checkpoint downloaded from https://dl.fbaipublicfiles.com/segment_anything/sam_vit_b_01ec64.pth
Put the downloaded checkpoint into pretrained/ folder
Go to preproccessing Folder and Create python environment using requirements.txt (Preprocessing uses different set of packages than trainig and must be done only once)
After Activating the newly created Environment run:
python run --tif-folder $ABSOLUTE_PATH_TO_WHERE_TIF_IMAGES_RESIDE --geojson-folder $ABSOLUTE_PATH_WHERE_GEOMETRY_MASKS_RESIDE
This will create and put resutls into ./temp folder
copy load folder from ./temp and put it into the root workspace path
load folder contains everython you need for training, validation, and testing
Go to workspace Folder and Create python environment using requirements.txt
Activate that workspace and run:
torchrun train.py --config configs/sam-vit-b.yaml
python test.py --model $ABSOLUTE_PATH_TO_THE_TRAINED_MODEL ----dataset ('val_dataset'|'test_dataset')