Dataset Viewer
The dataset viewer is not available for this dataset.
Unexpected token '<', "<html> <h"... is not valid JSON

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

标注仓库 README(模板)

本仓库用于管理标注过程数据与导出结果(非训练集本体)。

目录结构

  • meta/
    • classes.txt
    • label_schema.yaml
    • sources.yaml
    • splits.yaml
  • raw/
    • frames/
    • videos/
  • prelabel/
    • yolo/
  • tasks/
    • batch_0001/
      • images/
      • labels/
      • notes.md
  • curated/
    • hard_cases/
    • missing_candidates/
    • predictions/
    • pred_vis/
    • predict_image/
  • exports/
    • export_v001/
      • images/
      • labels/
      • export_meta.json

标注输入与输出

输入(标注前):

  • 待标注图片放入 tasks//images/
  • 预标注(可选)放入 prelabel/yolo/ 与图片同名

输出(标注后):

  • 标注结果(真值)输出到 tasks//labels/
  • 导出给训练集的结果输出到 exports//{images,labels}/

推荐最小流程

  1. 选样本:raw/frames -> tasks//images
  2. 预标注(可选):auto_label_yolo.py -> prelabel/yolo
  3. 人工标注:输出 tasks//labels
  4. 导出:复制 tasks/ 到 exports/
  5. 训练集仓库同步:exports/ -> train dataset

版本挂钩

  • 标注仓库版本:annot_v001(tag)
  • 训练集仓库引用:dataset_meta.json 中记录 source_annotation_revision
  • 导出版本:exports//export_meta.json

注意事项

  • images 与 labels 必须同名一一对应
  • 只在 exports/ 下形成“可训练的干净版本”
  • curated/ 仅用于分析与回溯,不进入训练集
Downloads last month
101