Setup

First, extract the zip files from the example_data directory and place them in the root directory of this project. This is the result of a run using the default parameters (gray_threshold=5, iou_threshold=0.50), the masks outputted by SegGPT, and the YOLO data files from a previous run that I took from the AMPLIfy SegGPT OneDrive location. Then you won't have to mess with any of the directory parameters for the scripts, they will default to the names of the folders extracted

Usage

Convert SegGPT masks to bounding boxes using OpenCV

python mask2boundingboxes.py
  --input_directory {directory of masks to draw bounding boxes around}
  --mask_output_directory {directory to output masks to with bounding boxes drawn}
  --yolo_output_directory {directory to output text files with the bounding boxes in YOLO format}
  --gray_threshold {the level of light (0-255) to detect as part of the mask and draw a bounding box around, default=5}
  --limit {a limit on the number of masks to attempt to process}

Analyze the accuracy of the bounding boxes drawn around the SegGPT masks, compared to a directory of YOLO text files from previous runs

python compare_results.py
  --seggpt_directory {directory containing the bounding box YOLO files outputted by mask2boundingboxes.py}
  --yolo_directory {directory containing the bounding box YOLO files to compare to}
  --output_directory {directory containing mask images (processed or unprocessed) to retrieve image dimensions from for denormalization of YOLO data}
  --iou_threshold {a float between 0-1 representing what percentage of the volume of a bounding box must overlap to consider "matching", default=0.50}

Manually draw bounding boxes on masks given existing YOLO data (for visual inspection)

python yolo2boundingboxes.py
  --yolo_directory {directory containing text files of bounding box data in YOLO format}
  --mask_directory {directory of masks to draw bounding boxes around}
  --output_directory {directory to output masks to with bounding boxes drawn}
  --limit {a limit on the number of masks to attempt to process}