# DMS 标注分期验收(addw pilot) ## P0:dam / ddaw(检测) - [ ] 批次台账入湖 → inbox 有 `images/train` - [ ] 送标工作台「待送标」可见批次 - [ ] 进入标注:左侧工具箱(face 等)+ 保存 - [ ] 提交后 `batch.meta` stage = `labeling_submitted`,「待入库」Tab 可见 - [ ] 导出 Job 成功后 stage = `returned` - [ ] 提交入库审核 → build 通过 ## P1:addw / addw_face(当前 pilot) Pilot:`inbox/addw/20260525_pilot`(12 张) - [x] `dms_detect.xml` 四类标签 - [x] `dms_pose.xml`:face 框 + 37 关键点(`kp_00`–`kp_36`) - [x] `export_ls_to_yolo.py`:LS JSON → YOLO detect / YOLO pose - [ ] 标注保存至 `labels/ls_annotations/*.json` - [ ] 提交 / 导出 / 待入库全链路(见 P0 检查项) - [ ] addw_face 导出 txt 每行 116 字段,通过 `validate_pose_label` ```bash python3 datasets/dms/scripts/test_export_ls_to_yolo.py ``` ## P2:forward / lane - [ ] `forward__detect` / `forward__classify` registry 模板 - [ ] lane `lane_culane.xml` 画布 smoke ## 自动化 ```bash bash scripts/smoke_labeling_api.sh bash scripts/smoke_pending_gate.sh python3 scripts/lake_checklist_audit.py ```