feat: 建立 SAM2 标注闭环基线

- 打通工作区真实标注闭环:支持手工多边形、矩形、圆形、点区域和线段生成 mask,并可保存、回显、更新和删除后端 annotation。

- 增强 polygon 编辑器:支持顶点拖动、顶点删除、边中点插入、多 polygon 子区域选择编辑,以及区域合并和区域去除。

- 接入 GT mask 导入:后端支持二值/多类别 mask 拆分、contour 转 polygon、distance transform seed point,前端支持导入、回显和 seed point 拖动编辑。

- 完善导出能力:COCO JSON 导出对齐前端,PNG mask ZIP 同时包含单标注 mask、按 zIndex 融合的 semantic_frame 和 semantic_classes.json。

- 打通异步任务管理:新增任务取消、重试、失败详情接口与 Dashboard 控件,worker 支持取消状态检查并通过 Redis/WebSocket 推送 cancelled 事件。

- 对接 Dashboard 后端数据:概览统计、解析队列和实时流转记录从 FastAPI 聚合接口与 WebSocket 更新。

- 增强 AI 推理参数:前端发送 crop_to_prompt、auto_filter_background 和 min_score,后端支持点/框 prompt 局部裁剪推理、结果回映射和负向点/低分过滤。

- 接入 SAM3 基础设施:新增独立 Python 3.12 sam3 环境安装脚本、外部 worker helper、后端桥接和真实 Python/CUDA/包/HF checkpoint access 状态检测。

- 保留 SAM3 授权边界:当前官方 facebook/sam3 gated 权重未授权时状态接口会返回不可用,不伪装成可推理。

- 增强前端状态管理:新增 mask undo/redo 历史栈、AI 模型选择状态、保存状态 dirty/draft/saved 流转和项目状态归一化。

- 更新前端 API 封装:补充 annotation CRUD、GT mask import、mask ZIP export、task cancel/retry/detail、AI runtime status 和 prediction options。

- 更新 UI 控件:ToolsPalette、AISegmentation、VideoWorkspace 和 CanvasArea 接入真实操作、导入导出、撤销重做、任务控制和模型状态。

- 新增 polygon-clipping 依赖,用于前端区域 union/difference 几何运算。

- 完善后端 schemas/status/progress:补充 AI 模型外部状态字段、任务 cancelled 状态和进度事件 payload。

- 补充测试覆盖:新增后端任务控制、SAM3 桥接、GT mask、导出融合、AI options 测试;补充前端 Canvas、Dashboard、VideoWorkspace、ToolsPalette、API 和 store 测试。

- 更新 README、AGENTS 和 doc 文档:冻结当前需求/设计/测试计划,标注真实功能、剩余 Mock、SAM3 授权边界和后续实施顺序。
This commit is contained in:
2026-05-01 15:26:25 +08:00
parent f020ff3b4f
commit 689a9ba283
48 changed files with 3280 additions and 176 deletions

View File

@@ -20,9 +20,11 @@ from services.frame_parser import (
upload_frames_to_minio,
)
from statuses import (
PROJECT_STATUS_PENDING,
PROJECT_STATUS_ERROR,
PROJECT_STATUS_PARSING,
PROJECT_STATUS_READY,
TASK_STATUS_CANCELLED,
TASK_STATUS_FAILED,
TASK_STATUS_RUNNING,
TASK_STATUS_SUCCESS,
@@ -31,6 +33,10 @@ from statuses import (
logger = logging.getLogger(__name__)
class TaskCancelled(RuntimeError):
"""Raised internally when a persisted task has been cancelled."""
def _now() -> datetime:
return datetime.now(timezone.utc)
@@ -66,12 +72,29 @@ def _set_task_state(
publish_task_progress_event(task)
def _project_status_after_stop(project: Project) -> str:
return PROJECT_STATUS_READY if project.frames else PROJECT_STATUS_PENDING
def _ensure_not_cancelled(db: Session, task: ProcessingTask) -> None:
db.refresh(task)
if task.status == TASK_STATUS_CANCELLED:
raise TaskCancelled("Task was cancelled")
def run_parse_media_task(db: Session, task_id: int) -> dict[str, Any]:
"""Parse one project's media and update task progress in the database."""
task = db.query(ProcessingTask).filter(ProcessingTask.id == task_id).first()
if not task:
raise ValueError(f"Task not found: {task_id}")
if task.status == TASK_STATUS_CANCELLED:
return {
"task_id": task.id,
"status": TASK_STATUS_CANCELLED,
"message": task.message or "任务已取消",
}
if task.project_id is None:
_set_task_state(
db,
@@ -111,6 +134,7 @@ def run_parse_media_task(db: Session, task_id: int) -> dict[str, Any]:
db.commit()
raise ValueError("Project has no media uploaded")
_ensure_not_cancelled(db, task)
project.status = PROJECT_STATUS_PARSING
_set_task_state(db, task, status=TASK_STATUS_RUNNING, progress=5, message="后台解析已启动", started=True)
@@ -121,6 +145,7 @@ def run_parse_media_task(db: Session, task_id: int) -> dict[str, Any]:
os.makedirs(output_dir, exist_ok=True)
try:
_ensure_not_cancelled(db, task)
_set_task_state(db, task, progress=15, message="正在下载媒体文件")
if effective_source == "dicom":
dcm_dir = os.path.join(tmp_dir, "dcm")
@@ -129,20 +154,24 @@ def run_parse_media_task(db: Session, task_id: int) -> dict[str, Any]:
client = get_minio_client()
objects = list(client.list_objects(BUCKET_NAME, prefix=project.video_path, recursive=True))
for obj in objects:
_ensure_not_cancelled(db, task)
if obj.object_name.lower().endswith(".dcm"):
data = download_file(obj.object_name)
local_dcm = os.path.join(dcm_dir, os.path.basename(obj.object_name))
with open(local_dcm, "wb") as f:
f.write(data)
_ensure_not_cancelled(db, task)
_set_task_state(db, task, progress=35, message="正在解析 DICOM 序列")
frame_files = parse_dicom(dcm_dir, output_dir)
else:
_ensure_not_cancelled(db, task)
media_bytes = download_file(project.video_path)
local_path = os.path.join(tmp_dir, Path(project.video_path).name)
with open(local_path, "wb") as f:
f.write(media_bytes)
_ensure_not_cancelled(db, task)
_set_task_state(db, task, progress=35, message="正在使用 FFmpeg/OpenCV 拆帧")
frame_files, original_fps = parse_video(local_path, output_dir, fps=int(parse_fps))
project.original_fps = original_fps
@@ -158,12 +187,15 @@ def run_parse_media_task(db: Session, task_id: int) -> dict[str, Any]:
except Exception as exc: # noqa: BLE001
logger.warning("Thumbnail extraction failed: %s", exc)
_ensure_not_cancelled(db, task)
_set_task_state(db, task, progress=70, message="正在上传帧到对象存储")
object_names = upload_frames_to_minio(frame_files, project.id)
_ensure_not_cancelled(db, task)
_set_task_state(db, task, progress=85, message="正在写入帧索引")
frames_out = []
for idx, obj_name in enumerate(object_names):
_ensure_not_cancelled(db, task)
local_frame = frame_files[idx]
try:
import cv2
@@ -203,6 +235,23 @@ def run_parse_media_task(db: Session, task_id: int) -> dict[str, Any]:
)
logger.info("Parsed %d frames for project_id=%s", len(frames_out), project.id)
return result
except TaskCancelled:
project.status = _project_status_after_stop(project)
task.status = TASK_STATUS_CANCELLED
task.progress = 100
task.message = task.message or "任务已取消"
task.error = task.error or "Cancelled by user"
task.finished_at = task.finished_at or _now()
db.commit()
db.refresh(task)
publish_task_progress_event(task)
logger.info("Parse task cancelled: task_id=%s project_id=%s", task.id, project.id)
return {
"task_id": task.id,
"project_id": project.id,
"status": TASK_STATUS_CANCELLED,
"message": task.message,
}
except Exception as exc: # noqa: BLE001
project.status = PROJECT_STATUS_ERROR
_set_task_state(