AGENT SKILL

training-data

Routes COCO/VOC annotation preparation, class and anchor config, dataset validation, and training workflows for tensorflow-yolov4-tflite.

License 许可
MIT
最近更新
2026-09-03
Stars
★ 351
Forks
27

01 前言 Front Matter

name
training-data
license
MIT
metadata
{ "disco-role": "operating" }
description
Routes COCO/VOC annotation preparation, class and anchor config, dataset validation, and training workflows for tensorflow-yolov4-tflite.
disable-model-invocation
true

02 资源文件 4

03 SKILL.md 正文

Training Data and Training

Use this sub-skill when the user needs to prepare annotations, validate class files, adapt COCO/VOC data, change train/test config, or run the repository's YOLOv3/YOLOv4 training loop.

Before changing data or training

  • Read references/data-formats.md to confirm the annotation line format and class-id expectations.
  • Use scripts/validate_annotation_line.py on a tiny sample before launching training or evaluation.
  • Confirm the target checkout root is the command working directory. Default paths in core.config.cfg are relative to that root.
  • Confirm class-file order, anchors, model family, and tiny/full choice before loading or converting weights.

Main routes

  1. Validate converted annotations: run the bundled validator and fix malformed lines before touching TensorFlow.
  2. COCO preparation: read references/workflows.md for the repo's JSON-to-pickle and pickle-to-annotation sequence.
  3. VOC preparation: read references/workflows.md for the XML-to-annotation flow.
  4. Training configuration: read references/workflows.md for cfg.TRAIN values, freeze stages, checkpoints, and scratch/transfer caveats.
  5. Training troubleshooting: use references/troubleshooting.md for path, class-count, NumPy, dataset, and weight-loading failures.

Validator examples

Validate one converted annotation line:

python sub-skills/training-data/scripts/validate_annotation_line.py \
  --classes data/classes/coco.names \
  --line "image.jpg 10,20,100,200,0"

Validate a file and require image paths to exist:

python sub-skills/training-data/scripts/validate_annotation_line.py \
  --classes data/classes/coco.names \
  --annotation-file data/dataset/val2017.txt \
  --check-images

Do not use the repository's source data/dataset/*.txt files as portable training data; they contain source-author absolute image paths and are best used only as format examples.

Handoff to other sub-skills

Stop conditions

Stop before downloading COCO, running full training, modifying config for custom class counts, or overwriting checkpoints unless the user explicitly approves the runtime, data, and output policy.

04 同仓库更多 Skills

VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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).

★ 351