dataset-tools
Guide SAHI COCO dataset creation, slicing, filtering, conversion, evaluation, error analysis, and visualization workflows.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- dataset-tools
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Guide SAHI COCO dataset creation, slicing, filtering, conversion, evaluation, error analysis, and visualization workflows.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/coco-cli-and-api.md8.6 KB
- 参考文档references/data-formats.md6.8 KB
- 参考文档references/troubleshooting.md6.6 KB
- 脚本scripts/coco_fixture_smoke.py7.6 KB
03 SKILL.md 正文
Dataset Tools
Use this sub-skill when a task is about SAHI dataset operations rather than model construction or per-image prediction. It covers COCO-style dataset creation, loading, slicing, filtering, merging, train/validation splitting, YOLO export, COCO evaluation/error analysis, and FiftyOne visualization.
Route here for
- Building or loading COCO datasets with
Coco,CocoImage,CocoAnnotation,CocoPrediction, and category definitions. - Slicing a COCO dataset with
sahi.slicing.slice_cocoorsahi coco slice. - Remapping/filtering categories, clipping boxes to image dimensions, keeping or dropping negative images, subsampling/upsampling, merging datasets, and train/validation splitting.
- Exporting COCO datasets to YOLO/Ultralytics layout with
sahi coco yoloor theCocoexport helpers. - Running COCO evaluation, error-analysis plots, or FiftyOne visualization when optional dependencies are available.
Route away
- Per-image prediction, sliced inference, model loading, or detector backend setup: use the sliced-inference and model-integrations sibling sub-skills.
- Low-level conversion between SAHI prediction/result objects and annotation objects: use the annotations-and-results sibling sub-skill.
Operating references
- Read COCO CLI and API for SAHI dataset APIs, CLI commands, required optional dependencies, and workflow recipes.
- Read Data formats and output layouts before writing or consuming COCO JSON, sliced datasets, YOLO exports, COCO result JSON, evaluation outputs, or FiftyOne inputs.
- Read Troubleshooting when loading fails, categories do not match, negative samples disappear, boxes are clipped or invalid, evaluation dependencies are missing, FiftyOne does not launch, or YOLO symlinks fail.
- For a dependency-light local sanity check, run
python scripts/coco_fixture_smoke.pyfrom this sub-skill directory. The script creates only temporary files and does not requirepycocotools, FiftyOne, network access, model weights, or training.
Minimal decision checklist
- Confirm whether the task needs API usage, CLI usage, or both.
- Confirm the image directory that
images[*].file_nameresolves against. - Decide whether images without annotations must be preserved
(
ignore_negative_samples=False) or excluded. - Decide whether category IDs must be remapped before splitting or YOLO export.
- Treat
pycocotoolsand FiftyOne as optional: guard those workflows and provide a fallback JSON validation path when they are unavailable.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).