AGENT SKILL

inference-evaluation

Run checkpoint-backed LaneNet inference, TuSimple batch evaluation, and postprocess troubleshooting.

License 许可
Apache-2.0
最近更新
2026-09-03
Stars
★ 351
Forks
27

01 前言 Front Matter

name
inference-evaluation
license
Apache 2.0
metadata
{ "disco-role": "operating" }
description
Run checkpoint-backed LaneNet inference, TuSimple batch evaluation, and postprocess troubleshooting.
disable-model-invocation
true

02 资源文件 5

03 SKILL.md 正文

Inference and Evaluation

Use this sub-skill when the task is to run or debug LaneNet checkpoint inference, TuSimple-style batch evaluation, DBSCAN/lane-fit postprocessing, or output interpretation.

Route here for

  • Single-image inference from a TensorFlow checkpoint using the image preprocessing and restore behavior expected by LaneNet.
  • Batch inference/evaluation over a TuSimple test_set/clips image tree and saving overlay outputs.
  • Diagnosing empty or black lane masks, DBSCAN clustering failures, missing remap files, checkpoint restore mismatches, and headless plotting issues.
  • Understanding binary segmentation, instance embedding, clustered mask, fitted lane parameters, and saved overlay outputs.

Route elsewhere

  • Dataset conversion, list files, masks, or TFRecord generation: data-preparation.
  • Training, checkpoint creation, TensorBoard, or trainer configuration: training.
  • Frozen TensorFlow PB or MNN/mobile export: model-export.

Required context before acting

  1. Confirm the user has a LaneNet repo checkout or equivalent source tree with lanenet_model/, local_utils/, config/, and the remap file available from the run working directory.
  2. Confirm a TensorFlow 1.x-compatible Python environment. TensorFlow 1.15 with CUDA is verified for this skill; CPU can run inference/evaluation functionally but is slower.
  3. Confirm the checkpoint path. Pretrained weights are not bundled. The checkpoint argument should normally be the TensorFlow checkpoint base path, not the .index or .data-* shard.
  4. For custom images, ask whether TuSimple lane fitting is appropriate. If not, use --with_lane_fit 0 and inspect raw clustered masks before tuning fitted curves.

Bundled references and scripts

Operating rules

  • Do not imply pretrained weights ship with the skill or the repo checkout; obtain a checkpoint from training or a user-supplied pretrained download.
  • Prefer noninteractive saving in automation. Use display only when the environment has a working GUI and blocking windows are acceptable.
  • Keep inference/evaluation runs from the repository root or pass --repo_root to the bundled scripts so repo-relative config and remap loading resolve correctly.
  • For custom data, first validate preprocessing, checkpoint compatibility, and binary segmentation before changing DBSCAN or lane-fit settings.

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