peoplepeople000/drone-detection

Drone Detection Sample Frames

This dataset contains the detections/ sample frames exported from Assignment 3 Task 1.

Contents

• —One row per saved frame that contains at least one drone detection
• —Embedded image bytes in a Hugging Face-compatible image column
• —Aggregated detection metadata for all boxes found in that frame

Schema

• —video_id: source video identifier
• —frame_index: sampled frame index within the extracted-frame sequence
• —frame_name: saved image filename
• —timestamp_sec: frame timestamp in the source video
• —image: struct with bytes and path
• —image_path: local path of the saved detection frame
• —width: frame width in pixels
• —height: frame height in pixels
• —num_detections: number of detections in the frame
• —max_confidence: highest confidence score in the frame
• —mean_confidence: mean confidence score across detections in the frame
• —detections: list of detection structs containing class/confidence/bounding box fields
• —detections_json: JSON copy of detections for quick inspection

Generation

The Parquet file is created from detections/detections.csv with:

python scripts/export_hf_detection_dataset.py \
  --detections_csv detections/detections.csv \
  --output_parquet detections/sample_frames.parquet \
  --copy_csv_summary