data-preparation
Prepares and validates PaddleDetection COCO, VOC, WIDER, MOT, keypoint, custom, semi-supervised, and sliced-image datasets.
- License 许可
- Apache-2.0
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- data-preparation
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Prepares and validates PaddleDetection COCO, VOC, WIDER, MOT, keypoint, custom, semi-supervised, and sliced-image datasets.
- disable-model-invocation
- true
02 资源文件 3
- 参考文档references/conversion-workflows.md1.9 KB
- 参考文档references/data-formats.md1.9 KB
- 参考文档references/troubleshooting.md1.5 KB
03 SKILL.md 正文
Data Preparation
Use this route for annotation conversion, dataset directory layout, label lists, custom readers, COCO/VOC/MOT/keypoint data, semi-supervised splits, or small-object slicing.
Workflow
- Decide the task and schema before touching files: detection/instance segmentation, keypoint, MOT, rotated detection, or industrial pipeline input.
- Choose the matching dataset config under the target checkout's
configs/datasets/and record its dataset root, image directory, annotation path, category/label source, metric, andnum_classes. - Validate a tiny sample with
scripts/validate_detection_dataset.pybefore downloading full data or starting workers. - For custom annotations, prefer COCO JSON or VOC XML. Use the documented converter command patterns in
references/conversion-workflows.md; keep the source data and generated annotations separate. - For MOT, verify
images/,labels_with_ids/, sequence metadata, normalized boxes, class IDs, and identity IDs. For keypoints, verify the keypoint order and visibility convention. - For semi-supervised or sliced datasets, make the output directory explicit and preserve the random seed, percent, slice size, overlap, and merge method.
Schema rules
- COCO detection boxes are
[x, y, width, height]; image IDs and category IDs must be consistent. - VOC XML boxes are corner coordinates
xmin, ymin, xmax, ymax; label names must match the label list/config. - MOT labels use
[class, identity, x_center, y_center, width, height]normalized to image dimensions;identity=-1means unknown. - COCO keypoints use triples
[x, y, visibility]in the model's expected order; missing/unlabeled points use the documented zero convention.
Read references/data-formats.md, references/conversion-workflows.md, and references/troubleshooting.md before launching training. Download helpers are network/large-data operations and are not run by the bundled validator.
04 同仓库更多 Skills
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
python-api
Programmatic Object-Detection-Metrics API guidance for BoundingBox, BoundingBoxes, Evaluator, enums, coordinate conversion, AP/mAP semantics, and source-style import diagnostics.
file-evaluation
Folder-based PASCAL VOC AP/mAP evaluation from ground-truth and detection text files.
object-detection-metrics
Guides legacy Object-Detection-Metrics PASCAL VOC AP/mAP evaluation from detection text folders and direct Python API classes.
nx-workspace-patterns
Configure and optimize Nx monorepo workspaces. Use when setting up Nx, configuring project boundaries, optimizing build caching, or implementing affected commands.