AGENT SKILL

targets-and-formats

Apply Albumentations pipelines safely to images, masks, bboxes, keypoints, multiple images, volumes, 3D masks, and labels without corrupting geometry or data formats.

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

01 前言 Front Matter

name
targets-and-formats
license
MIT
metadata
{ "disco-role": "operating" }
description
Apply Albumentations pipelines safely to images, masks, bboxes, keypoints, multiple images, volumes, 3D masks, and labels without corrupting geometry or data formats.
disable-model-invocation
true

02 资源文件 4

03 SKILL.md 正文

Targets and Formats

Use this sub-skill when a task involves Albumentations input/output keys, bbox or keypoint formats, label fields, additional targets, or 3D volume/mask targets. The central rule is: declare every geometric annotation contract in A.Compose(...) before calling the pipeline, then keep the data arrays and label arrays length-aligned after filtering.

Route First

  • Use references/data-formats.md to choose BboxParams, KeypointParams, bbox/keypoint formats, label fields, shape rules, and 3D target keys.
  • Use references/target-recipes.md for copy-adaptable recipes covering detection, keypoints, segmentation, additional targets, multi-image inputs, and volumes.
  • Use references/troubleshooting.md when a pipeline drops boxes/keypoints, rejects shapes, misreads YOLO boxes, or loses labels.
  • Use scripts/validate_targets.py for a small local preflight of bbox/keypoint JSON fixtures and optional volume/mask3d shapes before debugging a full training dataset.

Required Composition Contracts

  • Bboxes require bbox_params=A.BboxParams(...) whenever the call includes bboxes; keypoints require keypoint_params=A.KeypointParams(...) whenever the call includes keypoints.
  • BboxParams(format=...) accepts "coco", "pascal_voc", "albumentations", or "yolo"; Albumentations internally transforms bboxes as normalized [x_min, y_min, x_max, y_max].
  • KeypointParams(format=...) accepts "xy", "yx", "xya", "xys", "xyas", "xysa", or "xyz"; Albumentations internally transforms keypoints as [x, y, z, angle, scale].
  • Labels that must be filtered with bboxes/keypoints belong in label_fields, for example label_fields=["class_labels"] or label_fields=["keypoint_labels"].
  • 3D volumes use volume or volumes; 3D masks use mask3d or masks3d; shape checking expects D,H,W[,C] for single volumes and N,D,H,W[,C] for batches.

Safe Defaults

  • Use strict=True while developing to catch unknown input keys and invalid transform arguments early; route general Compose behavior to ../pipeline-composition/.
  • Keep is_check_shapes=True unless a dataset intentionally supplies nonmatching spatial targets and the code handles that explicitly.
  • For detection data with messy boxes, prefer clip=True, filter_invalid_bboxes=True and a conservative min_visibility or min_area rather than silently trusting source annotations.
  • For keypoints near crop borders, set remove_invisible=False only if downstream code can handle off-image coordinates; otherwise preserve label alignment via label_fields and accept filtering.
  • For tensor conversion and channel-order decisions, route to ../framework-integration/; for transform family selection, route to ../transform-catalog/; for save/load/replay, route to ../serialization-and-reproducibility/.

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