周翔宇 xzhou

0 Followers · 0 Following

• Joined on 2026-08-27

Repositories

1

Projects Packages Public Activity Starred Repositories

xzhou pushed to main at xzhou/ModelTrainEval 2026-09-15 17:26:24 +08:00

891f930de2 refactor: prepare_label_studio.py 的内容改动（补上 a3e08a9 漏掉的部分）

xzhou pushed to main at xzhou/ModelTrainEval 2026-09-15 17:26:07 +08:00

a3e08a9bbc refactor: prepare_yolo_dataset.py 改为 prepare_label_studio.py，产出放到 runs/

xzhou pushed to main at xzhou/ModelTrainEval 2026-09-15 17:19:56 +08:00

af0897f8c7 refactor(detect): 数据集改为扁平结构、不划分 train/val，Label Studio 文件放到根目录

xzhou pushed to main at xzhou/ModelTrainEval 2026-09-15 17:06:43 +08:00

1a43267b55 feat(detect): --label-studio 一步生成可导入 Label Studio 的复核视图

xzhou pushed to main at xzhou/ModelTrainEval 2026-09-15 16:46:00 +08:00

7adc5b6332 feat(detect): YOLO 数据集统一为 PNG、按类别重命名并附 classes.txt

xzhou pushed to main at xzhou/ModelTrainEval 2026-09-15 16:26:53 +08:00

cc053a6e9c fix(deps): add label-studio dependency in pyproject.toml

xzhou pushed to main at xzhou/ModelTrainEval 2026-09-15 14:49:24 +08:00

960f0d753a fix(configs): update training epochs in film.yaml and film.yaml for classifier and YOLO models

xzhou pushed to main at xzhou/ModelTrainEval 2026-09-11 16:34:17 +08:00

012b3efc73 feat(detect): 按来源类别处置无框图片（--empty-policy）

xzhou pushed to main at xzhou/ModelTrainEval 2026-09-11 14:34:18 +08:00

d6c6aca325 feat(detect): 默认模型换成 yolo26n

xzhou pushed to main at xzhou/ModelTrainEval 2026-09-11 14:29:29 +08:00

0e830fe253 fix(config): update data YAML path from boxes_gradcam to boxes_shap

xzhou pushed to main at xzhou/ModelTrainEval 2026-09-11 12:05:53 +08:00

8dc2fe6a2e feat(detect): 把弱监督框数据集链接到 data/ 并接上 YOLO 训练

xzhou pushed to main at xzhou/ModelTrainEval 2026-09-11 10:27:21 +08:00

86c86b105c refactor: 按实际的五步工作流裁掉无关代码

xzhou pushed to main at xzhou/ModelTrainEval 2026-09-10 16:26:58 +08:00

57446c2f10 refactor: 合并为单一的框生成入口，gradcam / shap 可选，默认全量

xzhou pushed to main at xzhou/ModelTrainEval 2026-09-10 16:10:53 +08:00

9aa87d31a6 feat(localization): SHAP Partition 定位方法，并把每图框数上限默认收紧为 1

xzhou pushed to main at xzhou/ModelTrainEval 2026-09-10 15:46:33 +08:00

fc7136fddf feat: 用 Grad-CAM 从分类模型生成弱监督框标签

xzhou pushed to main at xzhou/ModelTrainEval 2026-09-10 15:01:37 +08:00

33fdf396ee feat: 新增缺陷分类训练与评估链路

xzhou pushed to main at xzhou/ModelTrainEval 2026-09-10 12:04:11 +08:00

33d28dcc7c feat(cluster): clusters/ 默认改用硬链接，便于直接在 runs/ 下继续分子类

xzhou pushed to main at xzhou/ModelTrainEval 2026-09-10 11:26:05 +08:00

f8654f447b feat(export): 改为按 clusters/ 目录导出，并重定位失效的软链

xzhou pushed to main at xzhou/ModelTrainEval 2026-09-08 15:26:03 +08:00

9b31473b27 feat(export): 支持把原父目录名作为文件名前缀

xzhou pushed to main at xzhou/ModelTrainEval 2026-09-07 20:51:19 +08:00

6e543660ce feat(configs): 新增 resnet101 的自监督聚类配置