Object Detection: Dataset Conversion and Validation Scripts

5 scripts for converting, validating, and sampling object detection datasets, released by uv-scripts and updated in March 2026. These tools support 6 bounding box formats and allow users to process Hugging Face datasets without local cloning.

Use Cases

• Converting coco_xywh bounding boxes to yolo normalized format using the convert-hf-dataset.py script
• Limiting dataset processing to a specific subset using the --max-samples flag
• Validating the structure of object detection datasets hosted on the Hugging Face Hub

Strengths

• Supports 6 distinct bounding box formats including COCO and YOLO
• Includes 5 specialized scripts for conversion, validation, inspection, diffing, and sampling
• Requires no local cloning or setup when used with uv run

Limitations

• Contains no actual image or annotation data
• Functionality is dependent on the availability of external datasets on the Hugging Face Hub
• Requires the uv package manager for the simplified execution workflow

Provenance

Source

uv-scripts on Hugging Face

Freshness

Last updated March 2026