AGENT SKILL

openpcdet

Operate OpenPCDet 3D object detection workflows across runtime setup, dataset preparation, training/evaluation, inference/custom data, and model/config extension.

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

01 前言 Front Matter

name
openpcdet
license
Apache 2.0
metadata
{ "disco-role": "operating" }
description
Operate OpenPCDet 3D object detection workflows across runtime setup, dataset preparation, training/evaluation, inference/custom data, and model/config extension.
disable-model-invocation
true

02 资源文件 8

03 SKILL.md 正文

OpenPCDet Operating Skill

Use this skill when a task involves OpenPCDet, the pcdet Python package, OpenPCDet YAML configs, 3D point-cloud detection datasets, or OpenPCDet training/evaluation/demo workflows.

Route first

  • Need installation, CUDA extension, spconv, import, or environment diagnosis? Read sub-skills/runtime-and-ops/SKILL.md.
  • Need dataset layout, info pickle, ground-truth database, split, or DATA_CONFIG help? Read sub-skills/data-preparation/SKILL.md.
  • Need to train, resume, evaluate, run distributed jobs, or interpret outputs/checkpoints? Read sub-skills/training-and-evaluation/SKILL.md.
  • Need demo inference, custom point-cloud files, CustomDataset, visualization, or non-visual inference adaptation? Read sub-skills/inference-and-custom-data/SKILL.md.
  • Need config/model registry, YAML overrides, detector family selection, BEVFusion/MPPNet/CaDDN/VoxelNeXt/DSVT notes, or new model extension points? Read sub-skills/models-and-configs/SKILL.md.

Minimal public install and verification shape

For a fresh OpenPCDet checkout and a CUDA-capable Python environment, the package build shape is:

python -m pip install --no-build-isolation -e <OpenPCDet-checkout>
python scripts/inspect_openpcdet_runtime.py --repo <OpenPCDet-checkout> --require-cuda-ops

The build command is public and reproducible, but the exact CUDA/PyTorch/spconv variants must match the user's environment; read sub-skills/runtime-and-ops/SKILL.md before repairing a failing build.

Use bundled helpers instead of memorizing repository scripts

The skill bundles safe helper scripts under scripts/ and sub-skill scripts/. They do not run expensive dataset/training jobs unless explicitly asked.

  • scripts/inspect_openpcdet_runtime.py checks imports, PyTorch CUDA, spconv/cumm, optional packages, and compiled OpenPCDet ops.
  • scripts/summarize_openpcdet_config.py loads an OpenPCDet YAML config and summarizes dataset/model/optimization fields without building data loaders or models.
  • scripts/plan_openpcdet_command.py constructs train/test/demo/dataset-prep commands for an OpenPCDet checkout and only executes when --execute is supplied.
  • sub-skills/data-preparation/scripts/check_openpcdet_dataset_layout.py checks expected dataset folders and generated info/database products.
  • sub-skills/inference-and-custom-data/scripts/check_point_cloud_array.py validates .bin/.npy point-cloud shapes before demo/custom inference.
  • sub-skills/models-and-configs/scripts/inventory_openpcdet_configs.py inventories config YAMLs and detector/dataset registry names.

Operating stance

  1. Treat OpenPCDet as GPU-first. CPU-only imports do not prove model training, evaluation, or native op correctness.
  2. Keep config, dataset root, checkpoint, and model family aligned. Most runtime failures come from mixing a config with the wrong data layout, class list, or checkpoint.
  3. Prefer static/config/import probes before large jobs. Run full training/evaluation only after checking CUDA ops, spconv variant, dataset infos, ground-truth database products, and CLI command construction.
  4. For future work in another checkout, use checkout-relative paths supplied by the user or the current task. Do not rely on the source checkout used to construct this skill.

Core references

  • references/repo-provenance.md records source commit, package version, environment facts, and evidence paths.
  • references/repo-routing-metadata.json contains router placement metadata for repo-skill import tooling.
  • references/openpcdet-overview.md summarizes architecture, registries, and workflow entry points.
  • references/troubleshooting.md covers cross-cutting install/import/data/config/job failures.
  • references/source-script-map.md records which repository scripts were wrapped, adapted, or excluded.

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