Create and activate a new Conda environment with Python 3.11:

conda create -n cav-sam python=3.11
conda activate cav-sam

Install the necessary packages using pip:

pip install -r requirements.txt

Download the checkpoint of SAM2:

wget https://dl.fbaipublicfiles.com/segment_anything_2/072824/sam2_hiera_tiny.pt -O checkpoints/sam2_hiera_tiny.pt

2. Training LoRA

Execute the following command to start the LoRA training process on both the reference image and the target image:

python train_lora.py \
	--image-path "/path/to/your/image" \
	--lora-save-path "/path/to/save/lora" \
	--class-name "the_class_name_of_the_image"

The class name for different datasets:

• Chest X-Ray: chest X-ray
• Deepglobe: satellite image
• FSS-1000: seagull
• ISIC: dermoscopic lesion

3. Run Evaluation

After obtaining the trained LoRA weight files for the reference and target images, execute the following command to evaluate CAV-SAM:

python cav_sam.py 
    --reference-image-path "/path/to/your/reference/image" \
    --reference-mask-path "/path/to/your/reference/mask" \
    --target-image-path "/path/to/your/target/image" \
    --reference-lora-path "/path/to/save/reference/lora" \
    --target-lora-path "/path/to/save/target/lora" \
    --pred-save-path "/path/to/save/predictions" \
    --class-name "the_class_name_of_the_image"