Assignment_2_Cancer_Detection

Mammography Object Detection

Environment Setup

Create a virtual environment and install dependencies:

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Replace Transformers Library File

After installation, replace the modeling_grounding_dino.py file in the transformers library with the provided custom file:

# Find transformers installation path
python -c "import transformers; import os; print(os.path.dirname(transformers.__file__))"

# Copy the custom file (replace <TRANSFORMERS_PATH> with the path from above)
cp modeling_grounding_dino.py <TRANSFORMERS_PATH>/models/grounding_dino/modeling_grounding_dino.py

Training

Training is performed using Jupyter notebooks located in Train_Notebooks/:

• CoOp Training: coop.ipynb
• CoCoOp Training: cocoop.ipynb
• FixMatch Training: fixmatch.ipynb

Open and run the respective notebook for your desired training method.

Testing

CoOp Testing

python test_coop.py --dataset_root /path/to/dataset --model_path /path/to/models --n_ctx 4

Arguments:

• --dataset_root: Path to dataset directory (default: /mammography)
• --model_path: Path to trained model files (default: ./trained_models)
• --n_ctx: Number of learnable context tokens (default: 4)

CoCoOp Testing

python test_cocoop.py --dataset_root /path/to/dataset --model_path /path/to/models --n_ctx 4 --feature_level highest

Arguments:

• --dataset_root: Path to dataset directory (default: /mammography)
• --model_path: Path to trained model files (required)
• --n_ctx: Number of learnable context tokens (default: 4)
• --feature_level: Feature level - highest, lowest, or middle (required)

Zero-shot Testing

python zero_shot.py --dataset_root /path/to/dataset

Arguments:

• --dataset_root: Path to dataset directory (default: /mammography)