human-extractor

SKILL.md

Description GPU-accelerated pipeline for detecting, tracking, and classifying humans in dashcam footage. Processes MP4 videos to extract human crops with optional CLIP-based head covering classification, saving all outputs to a unified directory with comprehensive indexing.

Purpose Extract visual evidence of human presence from dashcam recordings for investigative analysis. Optimized for high throughput using NVDEC decoding, batched YOLOv8 detection, ByteTrack multi-object tracking, and optional CLIP classification.

| GPU Utilization | 80-90% | NVDEC + dual streams + large batches | | Throughput | 3-4 videos/min | Parking videos (2 FPS sampling) | | VRAM Usage | 6-10 GB | YOLO=64, CLIP=384 | | Latency | <30s per video | Including decode, detect, track, classify |

GPU-accelerated pipeline for detecting, tracking, and classifying humans in dashcam footage. This skill should be used when users need to extract human presence from videos, analyze dashcam footage for people detection, perform investigative analysis of human subjects in recordings, or classify individuals with optional head covering detection. Source: yousufjoyian/claude-skills.

View raw