script/README.md

Installation

• Install PyTorch
• Install required Python packages:

pip install datasets
pip install huggingface_hub
pip install ultralytics

Basic usage: Run the Filtering on WIT-base

Run the Filtering with Command-Line Arguments

python wit_filter.py --device cuda:0 --batch_size 32 --output_filtered_data_file_path /path/to/filtered_data_file.parquet

• --device: Set to "cpu" if GPU is unavailable (default: cuda:0)
• --batch_size: Adjust based on your available memory (default: 32)
• --output_filtered_data_file_path: Path to save the filtered results (default: filtered_data_file.parquet)

The filtered dataset will be saved at the path specified by --output_filtered_data_file_path.

Evaluation Mode Usage: Evaluate Detection Performance on WIT-base Subset

A curated evaluation subset of 30 WIT-base images is included to evaluate the detection model performance.

To enable evaluation mode and save filtered images into category-specific folders, use the --eval_mode flag and specify the image directory:

python wit_filter.py --device cuda:0 --batch_size 32 --output_filtered_data_file_path /path/to/filtered_data_file.parquet --eval_mode --filtered_image_dir path/to/image_filter_result_dir

• --eval_mode: Enable evaluation mode to save filtered images into category-specific folders
• --filtered_image_dir: Directory where the filtered images will be saved (default: image_filter_result_dir)

Filtered images will be organized into subfolders under filtered_image_dir:

• no_face/: No valid face detected
• valid_face_no_glasses/: Valid face detected, no glasses
• valid_face_with_eyeglasses/: Valid face with eyeglasses
• valid_face_with_sunglasses/: Valid face with sunglasses

Information about the Evaluation Data

📎 wit_eval_30.csv: Metadata for the evaluation set.

Column	Description
idx	Index in the original WIT-base dataset
has_face	0 = No face or too small, 1 = Valid face
glasses_type	0 = No glasses, 1 = Eyeglasses, 2 = Sunglasses

📎 data/: Directory containing all 30 images in the evaluation subset.