data-preparation
Guides SiamMask benchmark and training data acquisition, layout validation, crop/index preprocessing, VOT JSON generation, and data-path troubleshooting.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- data-preparation
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Guides SiamMask benchmark and training data acquisition, layout validation, crop/index preprocessing, VOT JSON generation, and data-path troubleshooting.
- disable-model-invocation
- true
02 资源文件 6
- 参考文档references/data-layouts.md2.5 KB
- 参考文档references/troubleshooting.md2.8 KB
- 参考文档references/workflows.md3.1 KB
- 脚本scripts/check_dataset_layout.py5.0 KB
- 脚本scripts/generate_vot_json.py4.6 KB
- 脚本scripts/run_data_prep.py4.4 KB
03 SKILL.md 正文
SiamMask Data Preparation
Use this sub-skill when the task is about VOT/DAVIS/YouTube-VOS benchmark data, COCO/DET/VID/YouTube-VOS training data, crop511 preprocessing, dataset JSON files, or missing data paths.
Safety First
Most original preprocessing steps involve large downloads, unzips, symlinks, multiprocessing crops, or generated files. Keep commands in dry-run/read-only mode until the user explicitly approves network, disk, and runtime costs.
Main Routes
| User intent | What to do |
|---|---|
| "Check whether data is prepared" | Run scripts/check_dataset_layout.py and read references/data-layouts.md. |
| "Generate VOT metadata JSON" | Use scripts/generate_vot_json.py on an existing VOT directory; it adapts the repo metadata generator without downloading data. |
| "Run original preprocessing/crop scripts" | Use scripts/run_data_prep.py in dry-run mode, then add --run only after approving side effects. |
| "Prepare training data" | Read references/workflows.md and validate COCO/DET/VID/YouTube-VOS generated outputs before training. |
| "Dataset errors during tracking/training" | Read references/troubleshooting.md and root troubleshooting. |
Bundled Helpers
scripts/check_dataset_layout.py: read-only JSON report for VOT, DAVIS, YouTube-VOS, COCO, DET, VID, or the full training data mix.scripts/generate_vot_json.py: self-contained VOT metadata generator for existing local VOT datasets.scripts/run_data_prep.py: dry-run-first launcher for checkout-local data-preparation scripts; useful for auditing cwd,PYTHONPATH, and arguments before heavy preprocessing.
Example read-only check:
python scripts/check_dataset_layout.py --data-root <siammask-checkout>/data --dataset training
Example VOT metadata generation:
python scripts/generate_vot_json.py \
--dataset-root <siammask-checkout>/data/VOT2019 \
--dataset-name VOT2019 \
--output <siammask-checkout>/data/VOT2019.json
Cross-Links
- After benchmark data exists, use ../tracking/SKILL.md for test/eval/tune runs.
- After training data exists, use ../training/SKILL.md for CUDA training dry-runs.
- Use root install/setup for Cython extension and pycocotools build details.
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).