pykitti
Guide Python workflows for loading, validating, and interpreting KITTI raw, odometry, and tracking data with pykitti.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- pykitti
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Guide Python workflows for loading, validating, and interpreting KITTI raw, odometry, and tracking data with pykitti.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/api-overview.md3.1 KB
- 参考文档references/repo-provenance.md1.5 KB
- 参考文档references/repo-routing-metadata.json337 B
- 参考文档references/troubleshooting.md3.5 KB
- 脚本scripts/check_install.py2.0 KB
03 SKILL.md 正文
pykitti
Use this skill when a task names pykitti, KITTI raw drives, odometry
benchmark sequences, Velodyne .bin scans, OXTS packets, KITTI calibration,
stereo camera streams, or KITTI tracking labels. It teaches the public
pykitti==0.3.1 Python surface and safe validation patterns; it does not fetch
large datasets or run GUI demos.
Install and inspect
Install the public package in the Python environment that will run the workflow:
python -m pip install pykitti
# Required by this release's eager tracking import:
python -m pip install opencv-python-headless
python -c "import pykitti; print(pykitti.raw, pykitti.odometry, pykitti.tracking)"
The declared runtime dependencies are NumPy, matplotlib, Pillow, and pandas.
OpenCV is not declared by the package metadata, but pykitti.__init__ imports
tracking.py, which imports cv2 eagerly. Read
troubleshooting when top-level import fails.
Use the install diagnostic for a read-only import,
version, dependency, and signature check from an arbitrary working directory.
Choose the route
| User request or data layout | Read next |
|---|---|
Date/drive tree such as 2011_09_26/<date>_drive_<drive>_sync, OXTS, raw cameras, raw Velodyne |
raw-data |
sequences/<sequence>/calib.txt, times.txt, optional poses/<sequence>.txt |
odometry |
label_02, detection/label rows, object IDs, DontCare, per-frame boxes |
tracking |
| Shared rotations, calibration parsing, image or Velodyne primitives | API overview, then the owning route |
| Installation, missing files, dependency, alignment, or compatibility failure | troubleshooting |
Shared operating rules
- Identify the KITTI distribution and exact root layout before constructing a loader. A raw drive tree is not an odometry sequence tree.
- Validate metadata and compare sensor counts before synchronized processing. pykitti discovers files with sorted globs but does not reconcile stream lengths or numeric frame IDs.
- Use
framesas positional indices into sorted lists. Validate indices and list lengths yourself; do not assume a filename filter or alignment repair. - Use generator properties for sequential access and
get_*methods for indexed access. Materialize one-pass generators when they must be reused. - Keep frame names explicit:
T_destination_originis a homogeneous 4x4 transform,K_camNis a 3x3 intrinsic matrix, and Velodyne rows are[x, y, z, reflectance]. - Keep dataset acquisition, archive extraction, plotting, and GUI operations outside this runtime route. Prefer a trusted external data process and the bundled local fixture smokes for deterministic checks.
Verification
Run the safe fixture smoke belonging to the selected route before using a real archive. These scripts create temporary tiny fixtures and make no network or GUI calls:
- Raw:
python sub-skills/raw-data/scripts/raw_fixture_smoke.py - Odometry:
python sub-skills/odometry/scripts/odometry_fixture_smoke.py - Labels:
python sub-skills/tracking/scripts/labels_fixture_smoke.py
Read repository provenance before deciding whether this skill matches a later checkout or needs refreshing. The source snapshot is intentionally separate from the review artifacts and local inspection environment.
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).