data-preparation
Prepare and validate UniAD nuScenes, CAN bus, map, info-PKL, and motion-anchor data layout without launching training or evaluation.
- License 许可
- Apache-2.0
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- data-preparation
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Prepare and validate UniAD nuScenes, CAN bus, map, info-PKL, and motion-anchor data layout without launching training or evaluation.
- disable-model-invocation
- true
02 资源文件 3
- 参考文档references/data-formats.md6.3 KB
- 参考文档references/data-preparation.md6.6 KB
- 参考文档references/troubleshooting.md6.7 KB
03 SKILL.md 正文
UniAD data-preparation router
Use this sub-skill when the user needs to prepare, validate, or troubleshoot the data assets required by UniAD on nuScenes:
- raw nuScenes layout under
data/nuscenes/, including CAN bus and map extensions; - temporal info PKLs under
data/infos/; - the stage-2/E2E motion-anchor file under
data/others/; - safe command synthesis for UniAD's nuScenes info generation workflow.
Do not launch training, evaluation, visualization, or model editing from here.
Route boundaries
- Training, evaluation, checkpoint, GPU, distributed, or SLURM launch questions: route to
training-evaluation. - Model/config architecture, task heads, queue lengths, plugin classes, or registry questions: route to
config-and-model-architecture. - Result pickle interpretation, rendering, videos, or visualization commands: route to
visualization-and-results.
Operating procedure
- Identify the intended split/version: full
v1.0trainval/test, trainval-only, test-only, orv1.0-mini. - Identify the intended UniAD stage: BEVFormer/stage-1 versus stage-2/E2E. Stage-2/E2E requires
data/others/motion_anchor_infos_mode6.pkl. - Validate the local layout with
scripts/check_uniad_data_layout.pybefore constructing train/eval commands elsewhere. - If info PKLs must be generated, synthesize the command with
scripts/build_data_command.py; do not run a full conversion unless the user has confirmed the raw nuScenes/CAN bus/map assets and runtime dependencies are present. - Warn users that generated PKLs can embed root-prefixed paths. If a generated PKL stores paths such as
data/nuscenes/...or absolute dataset paths, the active UniAD config may needdata_root = ""instead of the defaultdata_root = "data/nuscenes/".
Bundled references
references/data-preparation.md— layout, download choices, info-generation command semantics, and config path cautions.references/data-formats.md— expected PKL, CAN bus, map, raw-sensor, and motion-anchor contents.references/troubleshooting.md— symptoms and fixes for missing directories, missing PKLs, motion anchors, root-path mismatches, network/download limits, and split confusion.
Bundled scripts
scripts/check_uniad_data_layout.py— self-contained dry validation of the expected data layout and common PKL path-root risk.scripts/build_data_command.py— self-contained dry command renderer adapted from UniAD's data conversion wrapper semantics.
04 同仓库更多 Skills
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
python-api
Programmatic Object-Detection-Metrics API guidance for BoundingBox, BoundingBoxes, Evaluator, enums, coordinate conversion, AP/mAP semantics, and source-style import diagnostics.
file-evaluation
Folder-based PASCAL VOC AP/mAP evaluation from ground-truth and detection text files.
object-detection-metrics
Guides legacy Object-Detection-Metrics PASCAL VOC AP/mAP evaluation from detection text folders and direct Python API classes.
nx-workspace-patterns
Configure and optimize Nx monorepo workspaces. Use when setting up Nx, configuring project boundaries, optimizing build caching, or implementing affected commands.