AGENT SKILL

framework-integration

Integrate Albumentations with PyTorch-style datasets, tensor conversion, optional extras, and DataLoader reproducibility.

License 许可
MIT
最近更新
2026-09-03
Stars
★ 351
Forks
27

01 前言 Front Matter

name
framework-integration
license
MIT
metadata
{ "disco-role": "operating" }
description
Integrate Albumentations with PyTorch-style datasets, tensor conversion, optional extras, and DataLoader reproducibility.
disable-model-invocation
true

02 资源文件 4

03 SKILL.md 正文

Framework Integration

Use this sub-skill when a task involves PyTorch dataset integration, tensor outputs, optional framework extras, or errors importing ToTensorV2 / ToTensor3D.

Route Here For

  • Adding Albumentations to a torch.utils.data.Dataset.__getitem__ while keeping augmentation before tensor conversion.
  • Fixing image, mask, batch, or volume tensor shapes produced by ToTensorV2 or ToTensor3D.
  • Explaining pip install albumentations[pytorch], albumentations[hub], and albumentations[text] optional dependencies.
  • Debugging missing A.ToTensorV2, HWC/CHW confusion, mask transposition, Normalize ordering, dtype/range issues, or DataLoader worker reproducibility.

Start With These References

  • references/pytorch-integration.md: PyTorch dataset placement, ToTensorV2, ToTensor3D, shapes, dtypes, normalization, bboxes/keypoints, and worker seeding.
  • references/optional-dependencies.md: Optional extras, import checks, and safe install guidance for PyTorch, Hub, and text features.
  • references/troubleshooting.md: Symptom-to-fix playbooks for import, shape, dtype/range, mask, and reproducibility failures.
  • scripts/pytorch_dataset_template.py: A tiny self-contained template/checker for dataset-style Albumentations pipelines.

Boundaries

  • For Compose, ReplayCompose, strict, additional_targets, and seed mechanics beyond PyTorch workers, use ../pipeline-composition/.
  • For selecting transforms such as crops, flips, color, dropout, or text augmentations, use ../transform-catalog/.
  • For bbox/keypoint coordinate formats, labels, volumes, and target validation contracts, use ../targets-and-formats/.
  • Keep model training loops, loss functions, optimizers, schedulers, and framework-specific trainer code out of scope unless they affect augmentation placement or returned sample format.

Minimal Pattern

import albumentations as A
from albumentations.pytorch import ToTensorV2

train_transform = A.Compose(
    [
        A.Resize(256, 256),
        A.HorizontalFlip(p=0.5),
        A.Normalize(mean=(0.485, 0.456, 0.406), std=(0.229, 0.224, 0.225)),
        ToTensorV2(),
    ],
    strict=True,
)

sample = train_transform(image=image_np, mask=mask_np)
image_tensor = sample["image"]  # C,H,W torch.Tensor
mask_tensor = sample["mask"]    # H,W torch.Tensor unless transpose_mask=True

Checklist

  • Install PyTorch support explicitly when tensor transforms are needed: pip install "albumentations[pytorch]" or install compatible torch separately.
  • Keep Albumentations inputs as NumPy arrays in H,W,C images, H,W or H,W,C masks, and D,H,W[,C] volumes until the final tensor transform.
  • Put Normalize before ToTensorV2; ToTensorV2 does not scale or normalize values by itself.
  • Use transpose_mask=True only when channel-first masks are expected by downstream code.
  • Return bbox/keypoint values from Albumentations as metadata/list fields; ToTensorV2 converts image/mask targets, not bbox/keypoint lists.

04 同仓库更多 Skills

VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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).

★ 351