datamodules-and-tasks
Use for TorchGeo Lightning DataModules, CLI/config runs, task classes, losses, metrics, prediction/plot hooks, and training workflow wiring.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- datamodules-and-tasks
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Use for TorchGeo Lightning DataModules, CLI/config runs, task classes, losses, metrics, prediction/plot hooks, and training workflow wiring.
- disable-model-invocation
- true
02 资源文件 1
- 参考文档references/datamodules-and-tasks.md2.4 KB
03 SKILL.md 正文
TorchGeo datamodules and tasks
Use this sub-skill when wiring TorchGeo into Lightning training, validation, testing, prediction, CLI configs, or task modules.
DataModule operating facts
BaseDataModulewraps a dataset class,batch_size,num_workers, split datasets, split dataloaders, and Kornia augmentation hooks.prepare_data()may download data whendownload=True; it must not create split state.setup(stage)creates datasets/samplers and is called on every process.GeoDataModuleis forGeoDatasetclasses and setscollate_fn = stack_samples. It managesRandomPatchSampler,GriddedPatchSampler, and batch samplers for split-specific patch extraction.NonGeoDataModuleis for integer-indexed datasets and ordinary PyTorch dataloader shuffle/split semantics.on_after_batch_transfer()applies the split-specific augmentation after the batch reaches the device. This is where Kornia GPU augmentations belong.
Task classes
TorchGeo task modules are Lightning modules for common geospatial ML problems:
Classification: timm-backed image classification withweights,in_channels, task type, class/label counts, losses, and optional frozen backbone.SemanticSegmentation: SMP or TorchGeo FCN segmentation withmodel,backbone,weights,in_channels, multiclass/multilabel/binary settings, losses, and freeze options.- Other task modules cover change detection, detection, instance segmentation, regression, temporal/spatiotemporal tasks, MAE, MoCo, BYOL, SimCLR, and IO benchmark workflows.
Before changing a task, inspect its corresponding tests/tasks/test_<task>.py for constructor matrix, batch schema, metric expectations, and trainer smoke coverage.
CLI/config workflow
- The
torchgeoconsole script is declared astorchgeo.main:main. - Prefer JSONArgParse/Lightning config files for reproducible training. Explicitly name the datamodule, task/model, trainer, optimizer/scheduler parameters, and paths.
- For smoke checks, use small fake fixtures and
max_epochs=1, small batch sizes, and no checkpoint downloads.
Batch and mask handling
- Images are typically tensors shaped
(B, C, H, W); some spatiotemporal paths use(B, T, C, H, W)and may flatten or route to spatiotemporal tasks. - Segmentation masks may require dtype conversion for Kornia and restoration before loss/metrics.
- Detection and instance segmentation batches can contain variable-length boxes/masks; use the datamodule utility collate function used by the tests.
Read next
- reference for task-specific examples and verification candidates.
- Root troubleshooting for Lightning, Kornia, and optional dependency failures.
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).