feature-extraction
Route Torchreid feature extraction, model discovery/building, checkpoint loading, distance scoring, rank evaluation, re-ranking, model complexity, ranked-result visualization, and activation-map visualization.
- License 许可
- MIT
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- feature-extraction
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Route Torchreid feature extraction, model discovery/building, checkpoint loading, distance scoring, rank evaluation, re-ranking, model complexity, ranked-result visualization, and activation-map visualization.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/api-reference.md5.5 KB
- 参考文档references/metrics-and-reranking.md4.1 KB
- 参考文档references/troubleshooting.md3.1 KB
- 参考文档references/workflows.md4.2 KB
03 SKILL.md 正文
Feature Extraction
Use this sub-skill when the task mentions any of the following:
FeatureExtractor, embeddings, or extracting features from images, numpy arrays, or tensors- comparing query and gallery sets
show_avai_modelsorbuild_modelload_pretrained_weightsor local checkpoint loadingcompute_distance_matrix,evaluate_rank, orre_rankingcompute_model_complexityvisualize_ranked_resultsorvisrank- activation maps /
visactmap/return_featuremaps=True
Route elsewhere
- Training, data managers,
Engine.run, or dataset setup → training-evaluation - ONNX / OpenVINO / TFLite export → model-export
- DML, OSNet-AIN NAS, or PA-100K attribute-recognition projects are excluded long-tail gaps unless a future extension bundles those project sources
Start here
Bundled helpers
- scripts/feature_extraction_smoke.py — CPU synthetic build/load/extract/metric smoke with no downloads.
- scripts/compare_embeddings.py — local-weight query/gallery embedding and distance helper.
- scripts/visualize_actmap.py — activation-map wrapper with dry-run preview by default.
Core rules
- Prefer local checkpoint paths. Do not rely on automatic pretrained downloads in helper scripts.
FeatureExtractorreturns a tensor shaped(B, D)and accepts image paths, HWC numpy arrays, ortorch.Tensorinputs.compute_distance_matrixreturns squared Euclidean distances formetric='euclidean'.evaluate_rankconsumes NumPy distance matrices plus pid/camid arrays and raises if none of the query identities appear in the gallery.- Activation-map visualization only works when the model
forward()acceptsreturn_featuremaps=Trueand returns 4-D feature maps at eval time.
04 同仓库更多 Skills
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
python-api
Programmatic Object-Detection-Metrics API guidance for BoundingBox, BoundingBoxes, Evaluator, enums, coordinate conversion, AP/mAP semantics, and source-style import diagnostics.
file-evaluation
Folder-based PASCAL VOC AP/mAP evaluation from ground-truth and detection text files.
object-detection-metrics
Guides legacy Object-Detection-Metrics PASCAL VOC AP/mAP evaluation from detection text folders and direct Python API classes.
nx-workspace-patterns
Configure and optimize Nx monorepo workspaces. Use when setting up Nx, configuring project boundaries, optimizing build caching, or implementing affected commands.