OpenSkillIndex← back to search
claude-code

AutoResearchClaw cv-detection

Best practices for object detection tasks. Use when working on COCO, VOC, or detection architectures like YOLO and DETR.

install
source · Clone the upstream repo
git clone https://github.com/aiming-lab/AutoResearchClaw
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/aiming-lab/AutoResearchClaw "$T" && mkdir -p ~/.claude/skills && cp -r "$T/researchclaw/skills/builtin/domain/cv-detection" ~/.claude/skills/aiming-lab-autoresearchclaw-cv-detection && rm -rf "$T"
manifest: researchclaw/skills/builtin/domain/cv-detection/SKILL.md
source content

Object Detection Best Practice

Architecture families:

  • One-stage: YOLO (v5/v8), SSD, RetinaNet, FCOS
  • Two-stage: Faster R-CNN, Cascade R-CNN
  • Transformer: DETR, DINO, RT-DETR

Training recipe:

  • Use pre-trained backbone (ImageNet)
  • Multi-scale training and testing
  • IoU threshold: 0.5 for mAP50, 0.5:0.95 for mAP
  • Use FPN for multi-scale feature extraction
  • Focal loss for class imbalance in one-stage detectors

Standard benchmarks:

  • COCO val2017: ~37 mAP (Faster R-CNN R50), ~51 mAP (DINO Swin-L)
  • Pascal VOC: ~80 mAP50 (Faster R-CNN)