domain-computer-vision
Skill8 starsChanged 4 months agoMITView on GitHub
Use for computer-vision tasks on natural/general images: image classification, object detection (vehicles, faces, people, products), face recognition/verification, semantic/instance segmentation, keypoints/pose, OCR. Picks the best architecture per sub-task and the right transfer-learning + augmentation recipe. Triggers on 'detect objects/faces/vehicles', 'classify images', 'segment', 'bounding boxes', 'recognize faces', 'image dataset'.
In this repository22 other files it gives its agents.
- accuracy-improvement-loopSkill
- data-rigor-and-leakageSkill
- dataset-profilingSkill
- deployment-explainabilitySkill
- domain-3dSkill
- domain-anomaly-detectionSkill
- domain-audio-speechSkill
- domain-generativeSkill
- domain-graphSkill
- domain-medical-imagingSkill
- domain-multimodalSkill
- domain-nlp-llmSkill
- domain-recommenderSkill
- domain-reinforcement-learningSkill
- domain-tabularSkill
- domain-time-seriesSkill
- domain-videoSkill
- literature-reviewSkill
- ml-research-methodologySkill
- notebook-deliverySkill
- rigorous-evaluationSkill
- training-optimizationSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 27
Detection data: verify box format (xywh vs xyxy, normalized?), anchor/scale sizing, and NMS thresholds.
- Very specific: line 30
Improve accuracy: use accuracy-improvement-loop.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Computer Vision - Method Selection
Pick by sub-task, then apply training-optimization + rigorous-evaluation. Prefer pretrained backbones; train from scratch only with very large data.
Decision table
| Sub-task | Recommended (strong default, then heavier option) | Notes |
| Image classification | EfficientNet / ConvNeXt, then ViT/Swin (needs more data) | ConvNeXt/EffNet are strong, data-efficient. Match input resolution to the backbone. |
| Object detection | YOLO (v8/11) / RT-DETR, then DINO-DETR or Faster R-CNN | YOLO for speed/deploy; DETR-family for accuracy. Use mAP@IoU. |
| Face detection | RetinaFace / YOLO-face / MTCNN | Detection ≠ recognition. |
| Face recognition/verification | ArcFace / CosFace embeddings + similarity | Metric-learning, not softmax classification. Evaluate with verification (ROC/EER), gallery split by identity. |
| Semantic segmentation | U-Net / DeepLabv3+ / SegFormer | Dice/IoU. Watch class imbalance. |
| Instance segmentation | Mask R-CNN / YOLO-seg | |
| Promptable / zero-shot seg | SAM / SAM2 | Great for annotation & few-shot. |
| Keypoints / pose | HRNet / YOLO-pose / ViTPose | |
| OCR | detector + recognizer (DBNet + CRNN) / TrOCR / PaddleOCR |
Cross-cutting CV practice
- Transfer learning from ImageNet; two-phase fine-tune; discriminative LR (see
training-optimization). - Augmentation matched to the task: flips only if label-invariant (a horizontally-flipped digit/letter is different!); mild rotation/scale/color; MixUp/CutMix help classification but can hurt localization and fine textures.
- Resolution is a top accuracy lever - small objects need higher resolution or tiling/patching; don't downsample away the signal.
- Detection data: verify box format (xywh vs xyxy, normalized?), anchor/scale sizing, and NMS thresholds.
- Leakage: split by source/scene/identity, not by frame - consecutive frames or same-identity images must not cross splits (
data-rigor-and-leakage). - Explainability: Grad-CAM for classifiers; overlay predicted boxes/masks for detection/segmentation.
- Improve accuracy: use
accuracy-improvement-loop. For tiny objects in big images, consider tiling or weakly-supervised localization.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.