AGENT SKILL

evaluation-and-visualization

Routes Ultra-Fast-Lane-Detection checkpoint evaluation, TuSimple scoring, CULane evaluation, and demo visualization workflows.

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

01 前言 Front Matter

name
evaluation-and-visualization
license
MIT
metadata
{ "disco-role": "operating" }
description
Routes Ultra-Fast-Lane-Detection checkpoint evaluation, TuSimple scoring, CULane evaluation, and demo visualization workflows.
disable-model-invocation
true

02 资源文件 6

03 SKILL.md 正文

evaluation-and-visualization

Use this sub-skill when a task is about scoring a trained checkpoint, building or invoking the CULane evaluator, generating TuSimple metrics, or creating the demo video outputs from the repo.

Read this when

  • You need to run test.py on CULane or TuSimple.
  • You need to score precomputed TuSimple predictions.
  • You need to build or call the CULane evaluator binary.
  • You need to generate AVI visualizations with demo.py.

What this sub-skill owns

  • Evaluation command construction.
  • CULane and TuSimple output formats.
  • Demo output and visualization behavior.
  • CULane evaluator build/run caveats.
  • Checkpoint loading and prefix compatibility on evaluation.

What it does not own

  • Training loop mechanics: see training.
  • Dataset layout and conversion: see data-and-config.
  • TorchScript export and speed timing: see export-and-speed.

Start here

  • Read references/evaluation-workflows.md for the practical command patterns.
  • Read references/output-formats.md for CULane line files and TuSimple JSONL expectations.
  • Read references/visualization.md when the user wants to inspect the demo output video.
  • Read references/troubleshooting.md when evaluation or visualization fails.
  • Use scripts/score_tusimple_json.py to score an existing TuSimple prediction file.
  • Use scripts/run_culane_evaluator.sh as the safer evaluator launcher.

Typical flow

  1. Confirm the dataset family and checkpoint path.
  2. Load the checkpoint with the repo's compatible-state logic.
  3. Run the evaluation command or the scoring helper.
  4. If the user wants visual output, run the demo workflow separately.
  5. If CULane F-measure is required, build the evaluator binary first.

Caution points

  • The repo's evaluation scripts call .cuda() directly.
  • TuSimple evaluation expects raw_file, lanes, h_samples, and run_time in the prediction JSONL.
  • CULane scoring depends on a native evaluator binary built from evaluation/culane/.
  • Demo outputs are written as AVI files and the source script assumes the default repo output naming.

Reference and script links

  • references/evaluation-workflows.md for the command patterns and binary requirements.
  • references/output-formats.md for the file formats.
  • references/visualization.md for demo behavior and output placement.
  • references/troubleshooting.md for checkpoint, evaluator, and OpenCV issues.
  • scripts/run_culane_evaluator.sh for a parameterized evaluator wrapper.
  • scripts/score_tusimple_json.py for TuSimple JSONL scoring.

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