segmentation-workflows
Operate MedicalZooPytorch 3D segmentation model selection, training, checkpointing, TensorBoard logging, and inference/visualization workflows.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- segmentation-workflows
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Operate MedicalZooPytorch 3D segmentation model selection, training, checkpointing, TensorBoard logging, and inference/visualization workflows.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/model-overview.md4.3 KB
- 参考文档references/troubleshooting.md3.2 KB
- 参考文档references/workflows.md5.6 KB
- 脚本scripts/smoke_model_factory.py9.6 KB
- 脚本scripts/smoke_writer.py6.1 KB
03 SKILL.md 正文
Segmentation workflows
Use this sub-skill for the 3D segmentation branch of MedicalZooPytorch: model selection, model construction, training loops, checkpointing, TensorBoard logging, and inference/visualization.
What this sub-skill covers
- Model factory routing through
lib.medzoo.create_model(args). - The common model base in
BaseModelfor checkpoint save/restore and inference. - The bundled training loop in
lib.train.Trainerand the generic scaffold inlib.train.BaseTrainer. - TensorBoard and CSV logging through
lib.visual3D_temp.TensorboardWriter. - Non-overlap inference and 3D volume visualization helpers in
lib.visual3D_temp.viz. - 3D model families:
UNET3D,VNET,VNET2,DENSENET1,DENSENET2,DENSENET3,HYPERDENSENET,SKIPDENSENET3D,DENSEVOXELNET,RESNET3DVAE,RESNETMED3D, andHIGHRESNET.
Fast path
- Choose a model family and a valid channel count for that family.
- Build the model with
lib.medzoo.create_model(args). - Create the loss, loaders, and trainer.
- Train, checkpoint, and monitor scalars.
- Use the visualization helpers for patch-based inference and saved volumes.
Entry points
references/model-overview.mdfor the supported 3D model families and factory behavior.references/workflows.mdfor training, checkpointing, TensorBoard, and inference patterns.references/troubleshooting.mdfor known model- and workflow-level pitfalls.scripts/smoke_model_factory.pyfor a safe factory-and-forward smoke check.scripts/smoke_writer.pyfor a temp-directory TensorBoard writer smoke check.
Operating notes
- The factory uses uppercase model ids and asserts membership in its model list.
HYPERDENSENETselects the 2-channel or 3-channel variant frominChannels.DENSENET2andDENSENET3are multi-stream models; match the input channel count to the branch you want. The bundled factory smoke skipsDENSENET2by default because the source channel math is known to fail there.RESNETMED3Dis wired through the factory at depth 18; use the direct generator if you need a different depth.- Some forward methods return tuples. Unpack them before you compute losses or log shapes.
- The bundled factory smoke has a
--include-known-brokenflag if you need to reproduce theDENSENET2channel-math failure explicitly. - The legacy inference demo depends on a checkpoint and GPU-style tensor placement; treat it as reference-only.
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).