AGENT SKILL

ultra-fast-lane-detection

Routes Ultra-Fast-Lane-Detection workflows for data preparation, training, evaluation, export, and speed/deployment tasks.

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

01 前言 Front Matter

name
ultra-fast-lane-detection
license
MIT
metadata
{ "disco-role": "operating" }
description
Routes Ultra-Fast-Lane-Detection workflows for data preparation, training, evaluation, export, and speed/deployment tasks.
disable-model-invocation
true

02 资源文件 5

03 SKILL.md 正文

Ultra-Fast-Lane-Detection

Use this skill when the task is about the Ultra-Fast-Lane-Detection lane detection repo: preparing CULane or TuSimple data, training a model, evaluating checkpoints, visualizing outputs, exporting TorchScript, or benchmarking speed.

Start here

  • Read references/repo-provenance.md if you want to confirm this skill matches the current checkout or to decide whether refresh-repo-skill is needed.
  • Run scripts/check_environment.py when you need a quick local smoke of imports, config parsing, and CUDA availability.
  • Use the sub-skill that matches the task family; this root file is only a router.

Install and runtime notes

  • This repo is not packaged as an installable distribution.
  • Use a Python environment with PyTorch and torchvision compatible with your backend, then install the runtime requirements listed by the repo (requirements.txt plus the small optional helpers used by the scripts here).
  • The documented workflows assume the repository root is on PYTHONPATH when you run the scripts from a checkout or copied tree.
  • CUDA is required for the native training, evaluation, demo, and speed scripts because the source code calls .cuda() directly.
  • The CULane evaluator also needs a C++ toolchain and OpenCV C++ development files.

Route map

sub-skills/data-and-config/

Read this when the task is about dataset layout, TuSimple conversion, config overrides, row anchors, or any data_root/log_path problem.

sub-skills/training/

Read this for single-GPU or distributed training, checkpoints, resume/finetune behavior, losses, schedulers, and TensorBoard logging.

sub-skills/evaluation-and-visualization/

Read this for test.py, CULane/TuSimple scoring, demo AVI generation, and evaluator build/run questions.

sub-skills/export-and-speed/

Read this for TorchScript export, synthetic speed checks, camera/video speed notes, and LibTorch/OpenCV deployment caveats.

Common entry points

  • data-and-config owns the data preparation path and the configuration defaults.
  • training owns the optimization loop and checkpoint lifecycle.
  • evaluation-and-visualization owns benchmark outputs and visual inspection.
  • export-and-speed owns deployment-oriented export and throughput checks.

When to read references

  • references/workflow-map.md for a compact overview of the repo workflows and where each sub-skill routes them.
  • references/troubleshooting.md for cross-cutting failures such as missing dependencies, CUDA runtime issues, checkpoint mismatches, and dataset path mistakes.
  • references/repo-provenance.md for staleness checks and repo snapshot details.

Safe smoke check

If you only need to confirm the repo knowledge is usable, run the bundled smoke helper before deeper work:

python scripts/check_environment.py --repo-root . --device cuda

If CUDA is not available, rerun with --device cpu to verify the non-GPU import path and record the missing backend separately.

Sub-skill reminders

  • Do not expect the original repo docs to stay open in a later session; the sub-skills and bundled references should carry the needed details.
  • Do not use this root file for long API tables, detailed command catalogs, or troubleshooting matrices; those belong in the sub-skills' references.
  • Do not mix evaluation, export, and training advice in one route when a dedicated sub-skill exists.

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