visualization-and-submission
Guide NAVSIM camera, BEV, and LiDAR visualization plus safe two-stage submission pickle preparation and local validation.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- visualization-and-submission
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Guide NAVSIM camera, BEV, and LiDAR visualization plus safe two-stage submission pickle preparation and local validation.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/submission.md6.3 KB
- 参考文档references/troubleshooting.md8.3 KB
- 参考文档references/visualization.md5.2 KB
- 脚本scripts/validate_submission_metadata.py6.9 KB
03 SKILL.md 正文
Visualization and submission
Use this route for NAVSIM scene plots, sensor overlays, GIFs, or a locally
validated warmup/challenge submission.pkl.
Read this when
- the task mentions
plot_bev_frame, camera grids, annotations, LiDAR overlays, custom plots, or GIF rendering - a user needs the submission pickle schema or metadata checks
- the workflow is
warmup_two_stageor the authorized privateprivate_test_hard_two_stagechallenge path
Start here
- Read visualization.md for scene loading, plotting entry points, sensor prerequisites, and custom overlays.
- Read submission.md before creating a pickle; it defines the stage containers, required metadata, parity rules, and the private-data boundary.
- Read troubleshooting.md for dependency, path, Hydra, plotting, and submission failures.
- Run the bundled submission metadata validator
on a trusted local pickle:
python scripts/validate_submission_metadata.py --help, thenpython scripts/validate_submission_metadata.py path/to/submission.pkl.
Operating rules
- Visualization needs a loaded
Scene, sensor blobs, calibration, and a map API for map layers; API/help checks do not need local data. - Submission evaluation receives only
AgentInput. The selected agent must haverequires_scene=False; a privileged agent that needsSceneor annotations must stop before generation. - Validate locally, but do not download data, train, run a benchmark, or upload a Hugging Face model as part of this route. Stop and ask for explicit authorization before any external/private operation.
- Use the same split and metric-cache path for warmup local evaluation and compare parity only after the local inputs are present. A valid pickle does not prove server acceptance or score parity.
Ownership boundaries
This route owns plotting and pickle preparation/validation. Route installation, data layout, sensor contracts, agent implementation, and EPDMS scoring to the corresponding NAVSIM skills; use the links in the references when composing a larger workflow.
04 同仓库更多 Skills
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.
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.
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.
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.
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.
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).