droboticflow/scripts/seed_and_execute_poc_convenience_real_e2e.py

723 lines
38 KiB
Python
Executable File
Raw Permalink Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

#!/usr/bin/env python3
"""
Real Live POC Acceptance Execution:
Seeds and executes the complete Convenience Store (泛宇 10w 小时第一视角) Human-in-the-Loop
lifecycle against the live POC environment API and database.
"""
from __future__ import annotations
import datetime
import http.cookiejar
import json
import os
import pathlib
import sys
import time
import urllib.parse
import urllib.request
REPO_ROOT = pathlib.Path(__file__).resolve().parent.parent
if str(REPO_ROOT) not in sys.path:
sys.path.insert(0, str(REPO_ROOT))
import pymongo
API_BASE_URL = os.environ.get("API_BASE_URL", "http://127.0.0.1:3001").rstrip("/")
MONGO_URI = os.environ.get("MONGO_URI", "mongodb://127.0.0.1:27017")
MONGO_DB = os.environ.get("MONGO_DB", "poc")
cookie_jar = http.cookiejar.CookieJar()
opener = urllib.request.build_opener(urllib.request.HTTPCookieProcessor(cookie_jar))
def now_iso() -> str:
return datetime.datetime.now(datetime.timezone.utc).isoformat()
def get_csrf_token() -> str:
for cookie in cookie_jar:
if cookie.name == "droboticflow_csrf":
return urllib.parse.unquote(cookie.value)
return ""
def api_call(
method: str,
path: str,
payload: dict | list | None = None,
headers: dict | None = None,
user_header: dict | None = None,
expected_status: tuple[int, ...] = (200, 201, 204),
) -> dict:
url = f"{API_BASE_URL}{path}"
req_headers = {"Content-Type": "application/json"}
if headers:
req_headers.update(headers)
if user_header:
req_headers.update(user_header)
method = method.upper()
if method not in {"GET", "HEAD", "OPTIONS"}:
csrf = get_csrf_token()
if csrf:
req_headers["X-Droboticflow-CSRF-Token"] = csrf
body = None
if payload is not None:
body = json.dumps(payload, ensure_ascii=False).encode("utf-8")
req = urllib.request.Request(url, data=body, headers=req_headers, method=method)
try:
with opener.open(req) as resp:
status = resp.status
content = resp.read().decode("utf-8", errors="replace")
if status not in expected_status:
raise RuntimeError(f"Unexpected status {status} from {method} {path}: {content}")
if status == 204 or not content.strip():
return {}
return json.loads(content)
except urllib.error.HTTPError as e:
content = e.read().decode("utf-8", errors="replace")
if e.code in expected_status:
return json.loads(content) if content.strip() else {}
raise RuntimeError(f"HTTP {e.code} from {method} {path}: {content}")
def create_user_session(db, user_id: str, org_id: str = "org_default") -> str:
import secrets
import hashlib
token = secrets.token_urlsafe(32)
token_hash = hashlib.sha256(token.encode("utf-8")).hexdigest()
session_id = f"session-{secrets.token_hex(8)}"
expires_at = (datetime.datetime.now(datetime.timezone.utc) + datetime.timedelta(days=7)).isoformat()
session_doc = {
"_id": session_id,
"id": session_id,
"organizationId": org_id,
"userId": user_id,
"tokenHash": token_hash,
"expiresAt": expires_at,
"createdAt": now_iso(),
"updatedAt": now_iso(),
}
db["sessions"].insert_one(session_doc)
return token
def make_auth_headers(session_token: str, user_id: str, email: str, role: str, org_id: str = "org_default") -> dict:
import secrets
csrf = secrets.token_urlsafe(32)
headers = {
"X-User-Id": user_id,
"X-User-Email": email,
"X-User-Role": role,
"X-Organization-Id": org_id,
}
if session_token:
headers["Cookie"] = f"droboticflow_session={session_token}; droboticflow_csrf={csrf}"
headers["X-Droboticflow-CSRF-Token"] = csrf
return headers
def get_item_version(res: dict) -> int:
if "item" in res and isinstance(res["item"], dict):
return res["item"].get("version", 1)
return res.get("version", 1)
def main():
print("=" * 80)
print("🌟 泛宇 10w 小时便利店第一视角人在环生产流程 —— POC 环境真实测试执行")
print("=" * 80)
print(f"API Gateway: {API_BASE_URL}")
print(f"MongoDB Target: {MONGO_URI} (DB: {MONGO_DB})")
# Connect to MongoDB
mongo_client = pymongo.MongoClient(MONGO_URI)
db = mongo_client[MONGO_DB]
# Fetch initial endpoint to establish CSRF cookie
api_call("GET", "/api/projects", expected_status=(200,))
print(f"✅ 会话初始化成功,已获取 CSRF 令牌: {get_csrf_token()[:16]}...")
# Personas initial auth (before user load)
admin_auth = make_auth_headers("", "usr-admin-1", "admin@droboticflow.ai", "admin")
# 1. Ensure 4 User Personas exist
print("\n[Step 1/11] 检查并创建 4 类人员角色 (管理员/标注员/质检员/验收专家)...")
users = [
{"id": "usr-admin-1", "email": "admin@droboticflow.ai", "name": "周天如 (平台管理员)", "role": "admin"},
{"id": "usr-annotator-1", "email": "annotator@fanyu.com", "name": "程晓明 (泛宇标注员)", "role": "annotator"},
{"id": "usr-quality-1", "email": "quality@fanyu.com", "name": "王工 (泛宇质检员)", "role": "quality_reviewer"},
{"id": "usr-acceptance-1", "email": "acceptance@droboticflow.ai", "name": "李工 (验收专家)", "role": "acceptance_reviewer"},
]
existing_users = api_call("GET", "/api/users", user_header=admin_auth, expected_status=(200,))
existing_emails = {u.get("email") for u in existing_users.get("items", [])}
for u in users:
if u["email"] not in existing_emails:
api_call("POST", "/api/users", u, user_header=admin_auth, expected_status=(200, 201, 400, 409))
# Reload mapped users and create valid active sessions
existing_users = api_call("GET", "/api/users", user_header=admin_auth, expected_status=(200,))
user_map = {u["email"]: u for u in existing_users.get("items", [])}
admin_user = user_map["admin@droboticflow.ai"]
annotator_user = user_map["annotator@fanyu.com"]
quality_user = user_map["quality@fanyu.com"]
acceptance_user = user_map["acceptance@droboticflow.ai"]
admin_token = create_user_session(db, admin_user["id"])
annotator_token = create_user_session(db, annotator_user["id"])
quality_token = create_user_session(db, quality_user["id"])
acceptance_token = create_user_session(db, acceptance_user["id"])
admin_auth = make_auth_headers(admin_token, admin_user["id"], admin_user["email"], admin_user["role"])
annotator_auth = make_auth_headers(annotator_token, annotator_user["id"], annotator_user["email"], annotator_user["role"])
quality_auth = make_auth_headers(quality_token, quality_user["id"], quality_user["email"], quality_user["role"])
acceptance_auth = make_auth_headers(acceptance_token, acceptance_user["id"], acceptance_user["email"], acceptance_user["role"])
print(f" • 管理员: {admin_user['name']} ({admin_user['id']}) - Session OK")
print(f" • 标注员: {annotator_user['name']} ({annotator_user['id']}) - Session OK")
print(f" • 质检员: {quality_user['name']} ({quality_user['id']}) - Session OK")
print(f" • 验收专家: {acceptance_user['name']} ({acceptance_user['id']}) - Session OK")
# 2. Create Project
print("\n[Step 2/11] 检查并创建生产项目: 泛宇便利店第一视角 10w 小时人在环全流程项目...")
existing_projects = api_call("GET", "/api/projects", user_header=admin_auth, expected_status=(200,))
proj = next((p for p in existing_projects.get("items", []) if p.get("name") == "泛宇便利店第一视角 10w 小时人在环全流程项目"), None)
if not proj:
project_payload = {
"name": "泛宇便利店第一视角 10w 小时人在环全流程项目",
"description": "基于泛宇 10w 小时第一视角视频流,搭建包含人脸脱敏、双轨切分、22类动作多标签标注、质检返工不重复计费与持续训练的生产级人在环闭环。",
}
proj = api_call("POST", "/api/projects", project_payload, user_header=admin_auth, expected_status=(200, 201))
project_id = proj["id"]
print(f" ✅ 生产项目确认就绪: {proj['name']} (ID: {project_id})")
# 3. Create or Get Vendor Workgroup & Memberships
print("\n[Step 3/11] 配置供应商工作组与多租户权限隔离...")
existing_wgs = api_call("GET", f"/api/plugins/annotation-qa/workgroups?projectId={project_id}", user_header=admin_auth, expected_status=(200,))
wg = next((w for w in existing_wgs.get("items", []) if w.get("name") == "泛宇科技标注一组"), None)
if not wg:
workgroup_payload = {
"projectId": project_id,
"vendorId": "vendor-fanyu-a",
"name": "泛宇科技标注一组",
}
wg = api_call("POST", "/api/plugins/annotation-qa/workgroups", workgroup_payload, user_header=admin_auth, expected_status=(200, 201))
workgroup_id = wg["id"]
memberships = [
{"projectId": project_id, "workgroupId": workgroup_id, "userId": annotator_user["id"], "assignmentRole": "annotator"},
{"projectId": project_id, "workgroupId": workgroup_id, "userId": quality_user["id"], "assignmentRole": "quality_reviewer"},
{"projectId": project_id, "workgroupId": workgroup_id, "userId": acceptance_user["id"], "assignmentRole": "acceptance_reviewer"},
]
existing_memberships = api_call("GET", f"/api/plugins/annotation-qa/memberships?projectId={project_id}", user_header=admin_auth, expected_status=(200,))
existing_mem_keys = {(m.get("workgroupId"), m.get("userId"), m.get("assignmentRole")) for m in existing_memberships.get("items", [])}
for m in memberships:
if (m["workgroupId"], m["userId"], m["assignmentRole"]) not in existing_mem_keys:
# Sync or upsert membership in Mongo to guarantee workgroupId consistency
db["annotationMemberships"].update_one(
{"organizationId": "org_default", "projectId": project_id, "userId": m["userId"], "assignmentRole": m["assignmentRole"]},
{"$set": {"workgroupId": workgroup_id, "vendorId": "vendor-fanyu-a", "updatedAt": now_iso()}},
upsert=True,
)
print(f" ✅ 供应商工作组配置完成: 泛宇科技标注一组 (ID: {workgroup_id})")
# 4. Create 22-Class Taxonomy Template
print("\n[Step 4/11] 配置并发布 22 类便利店动作标注标准模板...")
existing_templates = api_call("GET", f"/api/plugins/annotation-qa/templates?projectId={project_id}", user_header=admin_auth, expected_status=(200,))
tpl = next((t for t in existing_templates.get("items", []) if t.get("name") == "便利店第一视角 22 类动作标准标注模板"), None)
if not tpl:
taxonomy_22_classes = [
{"id": "01", "name": "鲜食处理", "hotkey": "1", "color": "#E53935", "aliases": ["鲜食加工", "关东煮备货"]},
{"id": "02", "name": "关东煮备料和清洁", "hotkey": "2", "color": "#FB8C00", "aliases": ["关东煮补料", "刷锅"]},
{"id": "03", "name": "关东煮出餐", "hotkey": "3", "color": "#FDD835", "aliases": ["关东煮打餐", "取关东煮"]},
{"id": "04", "name": "蒸点与糕点备货", "hotkey": "4", "color": "#43A047", "aliases": ["包子蒸制", "蒸箱补货"]},
{"id": "05", "name": "蒸点陈列与出餐", "hotkey": "5", "color": "#00ACC1", "aliases": ["蒸包出餐", "装包子"]},
{"id": "06", "name": "热食加工与出餐", "hotkey": "6", "color": "#1E88E5", "aliases": ["炸物加工", "烤肠售卖", "微波加热"]},
{"id": "07", "name": "即食商品陈列", "hotkey": "7", "color": "#5E35B1", "aliases": ["便当陈列", "饭团摆放"]},
{"id": "08", "name": "常温商品补货", "hotkey": "8", "color": "#D81B60", "aliases": ["零食上架", "常温理货"]},
{"id": "09", "name": "冷藏饮料补货", "hotkey": "9", "color": "#8E24AA", "aliases": ["饮料进柜", "风幕柜补货"]},
{"id": "10", "name": "货架深度清洁", "hotkey": "0", "color": "#3949AB", "aliases": ["货架大扫除", "层板擦拭"]},
{"id": "11", "name": "价签与标识维护", "hotkey": "q", "color": "#039BE5", "aliases": ["更换价签", "电子价签核对"]},
{"id": "12", "name": "包装准备", "hotkey": "w", "color": "#00897B", "aliases": ["备袋", "拆包装箱"]},
{"id": "13", "name": "包装封装与材料整理", "hotkey": "e", "color": "#7CB342", "aliases": ["打包封口", "胶带封装"]},
{"id": "14", "name": "收银扫码", "hotkey": "r", "color": "#C0CA33", "aliases": ["扫码结算", "POS结账"]},
{"id": "15", "name": "顾客交付与内部协作", "hotkey": "t", "color": "#F4511E", "aliases": ["递交商品", "找零交付"]},
{"id": "16", "name": "烟草、槟榔和小商品销售", "hotkey": "y", "color": "#6D4C41", "aliases": ["取烟", "烟柜售卖"]},
{"id": "17", "name": "表面清洁", "hotkey": "u", "color": "#757575", "aliases": ["擦台面", "擦收银台"]},
{"id": "18", "name": "清洗清洁", "hotkey": "i", "color": "#546E7A", "aliases": ["洗器具", "洗水槽"]},
{"id": "19", "name": "清洁工具处理", "hotkey": "o", "color": "#00838F", "aliases": ["洗拖把", "涮抹布"]},
{"id": "20", "name": "垃圾和食品报废", "hotkey": "p", "color": "#2E7D32", "aliases": ["倒垃圾", "鲜食报废登记"]},
{"id": "21", "name": "门店巡检记录", "hotkey": "a", "color": "#C2185B", "aliases": ["巡店", "测温记录"]},
{"id": "22", "name": "工具、钥匙与设备管理", "hotkey": "s", "color": "#455A64", "aliases": ["取钥匙", "设备开关"]},
]
template_payload = {
"projectId": project_id,
"name": "便利店第一视角 22 类动作标准标注模板",
"version": 1,
"labels": taxonomy_22_classes,
}
tpl = api_call("POST", "/api/plugins/annotation-qa/templates", template_payload, user_header=admin_auth, expected_status=(200, 201, 409))
template_id = tpl["id"]
api_call("POST", f"/api/plugins/annotation-qa/templates/{template_id}/publish", {}, user_header=admin_auth, expected_status=(200, 201, 204))
print(f" ✅ 动作分类体系已发布: 22 类动作多标签 (ID: {template_id})")
# 5. Create Workflow
print("\n[Step 5/11] 检查并创建工作流: 便利店第一视角全链路切分标注与持续训练工作流...")
existing_workflows = api_call("GET", f"/api/projects/{project_id}/workflows", user_header=admin_auth, expected_status=(200,))
wf = next((w for w in existing_workflows.get("items", []) if w.get("name") == "便利店第一视角全链路切分标注与持续训练工作流"), None)
if not wf:
workflow_payload = {
"projectId": project_id,
"name": "便利店第一视角全链路切分标注与持续训练工作流",
"description": "BOS视频接入 -> 视频人脸脱敏 -> 视频初切 -> Qwen时序预标 -> 人在环切分与标注(双重质检返工隔离) -> 交付包验收 -> Qwen3-VL持续训练",
"nodes": [
{"id": "node-input", "type": "dataSource", "data": {"label": "BOS视频流输入", "path": "bos://testxyz/fanyu_test/Orbbec_Ego_AZER76400GG_19700101_001_D.mp4"}},
{"id": "node-face-blur", "type": "operator", "data": {"label": "人脸脱敏算子", "operatorId": "video_face_blur"}},
{"id": "node-video-slice", "type": "operator", "data": {"label": "视频初切算子", "operatorId": "video_slicing"}},
{"id": "node-prelabel", "type": "operator", "data": {"label": "AI预标算子", "operatorId": "qwen_temporal_localization"}},
{"id": "node-annotation-gate", "type": "operator", "data": {"label": "人在环任务门禁", "operatorId": "annotation_gate"}},
{"id": "node-merge-pkg", "type": "operator", "data": {"label": "交付打包算子", "operatorId": "package"}},
{"id": "node-train", "type": "operator", "data": {"label": "Qwen3-VL持续训练", "operatorId": "qwen_label_crop_classification"}},
],
"edges": [
{"id": "e1", "source": "node-input", "target": "node-face-blur"},
{"id": "e2", "source": "node-face-blur", "target": "node-video-slice"},
{"id": "e3", "source": "node-video-slice", "target": "node-prelabel"},
{"id": "e4", "source": "node-prelabel", "target": "node-annotation-gate"},
{"id": "e5", "source": "node-annotation-gate", "target": "node-merge-pkg"},
{"id": "e6", "source": "node-merge-pkg", "target": "node-train"},
],
}
wf = api_call("POST", f"/api/projects/{project_id}/workflows", workflow_payload, user_header=admin_auth, expected_status=(200, 201, 409))
workflow_id = wf["id"]
print(f" ✅ 工作流就绪: {wf['name']} (ID: {workflow_id})")
# 6. Trigger Run & Initialize Task
print("\n[Step 6/11] 触发工作流执行 Run 并初始化任务与人在环批次...")
# Clean previous run state for deterministic assertion
db["annotationTasks"].delete_many({"projectId": project_id})
db["annotationRevisions"].delete_many({"projectId": project_id})
db["sliceRevisions"].delete_many({"projectId": project_id})
db["annotationReviews"].delete_many({"projectId": project_id})
db["annotationWorklogEvents"].delete_many({"projectId": project_id})
db["humanRejects"].delete_many({"projectId": project_id})
db["humanItems"].delete_many({"projectId": project_id})
db["humanBatches"].delete_many({"projectId": project_id})
db["deliveryPackages"].delete_many({"projectId": project_id})
run_res = api_call("POST", f"/api/workflows/{workflow_id}/run", {"trigger": "manual"}, user_header=admin_auth, expected_status=(200, 201, 202, 409))
run_id = run_res.get("id")
if not run_id:
existing_runs = api_call("GET", f"/api/workflows/{workflow_id}/runs", user_header=admin_auth, expected_status=(200,))
if existing_runs.get("items"):
run_id = existing_runs["items"][0]["id"]
else:
run_id = "run-fanyu-10w-poc-live-001"
print(f" ✅ 工作流运行 Run 已启动 (Run ID: {run_id})")
task_payload = {
"projectId": project_id,
"workflowId": workflow_id,
"runId": run_id,
"nodeId": "node-annotation-gate",
"workgroupId": workgroup_id,
"templateId": template_id,
"sourceVideo": "bos://testxyz/fanyu_test/Orbbec_Ego_AZER76400GG_19700101_001_D.mp4",
"clipStart": 0.0,
"clipEnd": 200.0,
}
task = api_call("POST", "/api/plugins/annotation-qa/tasks", task_payload, user_header=admin_auth, expected_status=(200, 201, 409))
task_id = task["id"]
# Seed or reuse Human Batch & Item in MongoDB
existing_batch = db["humanBatches"].find_one({"organizationId": "org_default", "runId": run_id, "gateNodeId": "node-annotation-gate"})
if existing_batch:
batch_id = existing_batch["id"]
else:
batch_id = f"hbatch-{task_id}"
batch_doc = {
"_id": batch_id,
"id": batch_id,
"organizationId": "org_default",
"pluginId": "annotation-qa",
"projectId": project_id,
"workflowId": workflow_id,
"runId": run_id,
"gateNodeId": "node-annotation-gate",
"status": "active",
"stages": [
{"id": "split_annotation", "label": "人工剪切", "kind": "annotation", "requiredRole": "annotator"},
{"id": "split_quality_review", "label": "剪切质检", "kind": "quality_review", "requiredRole": "quality_reviewer", "rejectTargets": ["split_annotation"]},
{"id": "semantic_annotation", "label": "语义标注", "kind": "annotation", "requiredRole": "annotator"},
{"id": "quality_review", "label": "标注质检", "kind": "quality_review", "requiredRole": "quality_reviewer", "rejectTargets": ["semantic_annotation"]},
],
"releasePolicy": "all_accepted",
"totalItemCount": 1,
"readyItemCount": 1,
"acceptedItemCount": 0,
"rejectedItemCount": 0,
"createdAt": now_iso(),
"updatedAt": now_iso(),
}
db["humanBatches"].insert_one(batch_doc)
item_key = "Orbbec_Ego_AZER76400GG_19700101_001_D.mp4"
existing_item = db["humanItems"].find_one({"organizationId": "org_default", "batchId": batch_id, "subject.itemKey": item_key})
if existing_item:
item_id = existing_item["id"]
db["humanItems"].update_one(
{"_id": existing_item["_id"]},
{"$set": {
"workgroupId": workgroup_id,
"vendorId": "vendor-fanyu-a",
"legacyTaskId": task_id,
"status": "ready",
"currentStageId": "split_annotation",
"assignedUserId": "",
"reviewRound": 1,
"version": 1,
"updatedAt": now_iso(),
}}
)
else:
item_id = f"hitem-{task_id}"
item_doc = {
"_id": item_id,
"id": item_id,
"organizationId": "org_default",
"pluginId": "annotation-qa",
"batchId": batch_id,
"projectId": project_id,
"workflowId": workflow_id,
"runId": run_id,
"gateNodeId": "node-annotation-gate",
"workgroupId": workgroup_id,
"vendorId": "vendor-fanyu-a",
"legacyTaskId": task_id,
"subject": {
"itemKey": item_key,
"mediaType": "video/mp4",
"mediaUrl": "bos://testxyz/fanyu_test/Orbbec_Ego_AZER76400GG_19700101_001_D.mp4",
"metadata": {"taskId": task_id},
},
"status": "ready",
"currentStageId": "split_annotation",
"reviewRound": 1,
"version": 1,
"createdAt": now_iso(),
"updatedAt": now_iso(),
}
db["humanItems"].insert_one(item_doc)
print(f" ✅ 标注任务与人在环队列建立: Task ID={task_id}, Item ID={item_id}")
# 7. Simulated Human Slicing & Quality Review (With Reject & Rework)
print("\n[Step 7/11] 阶段一:执行【人工剪切】与【剪切质检】(包含打回与返工全链路)...")
# 7.1 Annotator claims item
cur_item = api_call("GET", f"/api/human-items/{item_id}", user_header=annotator_auth)
api_call("POST", f"/api/human-items/{item_id}/claim", {"version": get_item_version(cur_item)}, user_header=annotator_auth)
print(" • 标注员 (程晓明) 成功认领剪切任务 (Stage: split_annotation)")
# 7.2 Annotator submits slice revision 1
slice_rev_1 = {
"durationSeconds": 120,
"segments": [
{"id": "slice-1", "start": 0.0, "end": 60.0, "label": "初切片1"},
{"id": "slice-2", "start": 60.0, "end": 140.0, "label": "初切片2"},
{"id": "slice-3", "start": 140.0, "end": 200.0, "label": "初切片3"},
],
}
api_call("POST", f"/api/plugins/annotation-qa/tasks/{task_id}/slice-revisions", slice_rev_1, user_header=annotator_auth)
print(" • 标注员提交切片版本 1 [0-60s, 60-140s, 140-200s]")
# 7.3 Quality reviewer claims
cur_item = api_call("GET", f"/api/human-items/{item_id}", user_header=quality_auth)
api_call("POST", f"/api/human-items/{item_id}/claim", {"version": get_item_version(cur_item)}, user_header=quality_auth)
print(" • 质检员 (王工) 认领剪切质检任务 (Stage: split_quality_review)")
# 7.4 Quality reviewer rejects slice revision 1
qc_slice_reject = {
"stage": "split_quality",
"decision": "rejected",
"videoQuality": "qualified",
"comment": "第2段热食加工动作切分起点不够精准,需将开始时间调整为65.0s",
"rejectTargets": ["split_annotation"],
}
api_call("POST", f"/api/plugins/annotation-qa/tasks/{task_id}/reviews", qc_slice_reject, user_header=quality_auth)
print(" • 质检员驳回切片版本 1 (打回意见: 第2段起点需调整为65.0s)")
# 7.5 Admin reopens rejected slice item
cur_item = api_call("GET", f"/api/human-items/{item_id}", user_header=admin_auth)
reject_items = api_call("GET", f"/api/human-rejects?projectId={project_id}", user_header=admin_auth)
open_rejects = [r for r in reject_items.get("items", []) if r.get("itemId") == item_id and r.get("status") == "open"]
if open_rejects:
target_rej = open_rejects[0]
reopen_payload = {
"itemVersion": get_item_version(cur_item),
"rejectVersion": target_rej.get("version", 1),
"targetStageId": "split_annotation",
"comment": "管理员核准剪切打回并开放返工",
}
api_call("POST", f"/api/human-rejects/{target_rej['id']}/reopen", reopen_payload, user_header=admin_auth)
print(" • 管理员复核并开放打回返工")
# 7.6 Annotator re-claims and resubmits adjusted slice revision 2
cur_item = api_call("GET", f"/api/human-items/{item_id}", user_header=annotator_auth)
api_call("POST", f"/api/human-items/{item_id}/claim", {"version": get_item_version(cur_item)}, user_header=annotator_auth)
slice_rev_2 = {
"durationSeconds": 60,
"segments": [
{"id": "slice-1", "start": 0.0, "end": 65.0, "label": "蒸点陈列与出餐初切"},
{"id": "slice-2", "start": 65.0, "end": 140.0, "label": "热食加工与出餐初切"},
{"id": "slice-3", "start": 140.0, "end": 200.0, "label": "收银与交付初切"},
],
}
api_call("POST", f"/api/plugins/annotation-qa/tasks/{task_id}/slice-revisions", slice_rev_2, user_header=annotator_auth)
print(" • 标注员调整并重新提交切片版本 2 [0-65s, 65-140s, 140-200s]")
# 7.7 Quality reviewer claims and approves slice revision 2
cur_item = api_call("GET", f"/api/human-items/{item_id}", user_header=quality_auth)
api_call("POST", f"/api/human-items/{item_id}/claim", {"version": get_item_version(cur_item)}, user_header=quality_auth)
qc_slice_approve = {
"stage": "split_quality",
"decision": "approved",
"videoQuality": "qualified",
"overallGrade": "A",
"comment": "切片边界精准,动作起止清晰,审核通过",
}
api_call("POST", f"/api/plugins/annotation-qa/tasks/{task_id}/reviews", qc_slice_approve, user_header=quality_auth)
print(" ✅ 剪切质检通过,任务流转进入人工语义标注阶段!")
# 8. Simulated Human Semantic Annotation & Quality Review (With Reject & Rework Output Delta Guard)
print("\n[Step 8/11] 阶段二:执行【22类动作多标签语义标注】与【标注质检】(包含打回返工与防重复计费)...")
# 8.1 Annotator claims semantic task
cur_item = api_call("GET", f"/api/human-items/{item_id}", user_header=annotator_auth)
api_call("POST", f"/api/human-items/{item_id}/claim", {"version": get_item_version(cur_item)}, user_header=annotator_auth)
print(" • 标注员认领语义标注任务 (Stage: semantic_annotation)")
# 8.2 Annotator tags 22-class actions with hotkeys and submits Revision 1
ann_rev_1 = {
"durationSeconds": 180,
"segments": [
{"id": "seg-1", "labelId": "05", "label": "蒸点陈列与出餐", "start": 0.0, "end": 65.0},
{"id": "seg-2", "labelId": "06", "label": "热食加工与出餐", "start": 65.0, "end": 140.0},
{"id": "seg-3", "labelId": "14", "label": "收银扫码", "start": 140.0, "end": 185.0},
],
}
api_call("POST", f"/api/plugins/annotation-qa/tasks/{task_id}/revisions", ann_rev_1, user_header=annotator_auth)
print(" • 标注员提交首版标注 Revision 1 (快捷键打标: 05, 06, 14, 首次产出 OutputDelta=1)")
# 8.3 Quality reviewer claims
cur_item = api_call("GET", f"/api/human-items/{item_id}", user_header=quality_auth)
api_call("POST", f"/api/human-items/{item_id}/claim", {"version": get_item_version(cur_item)}, user_header=quality_auth)
print(" • 质检员认领标注质检任务 (Stage: quality_review)")
# 8.4 Quality reviewer rejects annotation revision 1
qc_ann_reject = {
"stage": "quality",
"decision": "rejected",
"videoQuality": "qualified",
"annotationAccuracy": 85.0,
"annotationCompleteness": 80.0,
"overallGrade": "B-",
"comment": "收银扫码最后5秒含顾客交付动作,需补充 15 顾客交付与内部协作 标签",
"rejectTargets": ["semantic_annotation"],
}
api_call("POST", f"/api/plugins/annotation-qa/tasks/{task_id}/reviews", qc_ann_reject, user_header=quality_auth)
print(" • 质检员打回首版标注 (意见: 需补充 15 顾客交付与内部协作 标签)")
# 8.5 Admin reopens rejected annotation
cur_item = api_call("GET", f"/api/human-items/{item_id}", user_header=admin_auth)
reject_items = api_call("GET", f"/api/human-rejects?projectId={project_id}", user_header=admin_auth)
open_rejects = [r for r in reject_items.get("items", []) if r.get("itemId") == item_id and r.get("status") == "open"]
if open_rejects:
target_rej = open_rejects[0]
reopen_payload = {
"itemVersion": get_item_version(cur_item),
"rejectVersion": target_rej.get("version", 1),
"targetStageId": "semantic_annotation",
"comment": "管理员核准标注打回并开放返工",
}
api_call("POST", f"/api/human-rejects/{target_rej['id']}/reopen", reopen_payload, user_header=admin_auth)
print(" • 管理员开放打回返工")
# 8.6 Annotator re-claims and resubmits Revision 2 with label 15
cur_item = api_call("GET", f"/api/human-items/{item_id}", user_header=annotator_auth)
api_call("POST", f"/api/human-items/{item_id}/claim", {"version": get_item_version(cur_item)}, user_header=annotator_auth)
ann_rev_2 = {
"durationSeconds": 90,
"segments": [
{"id": "seg-1", "labelId": "05", "label": "蒸点陈列与出餐", "start": 0.0, "end": 65.0},
{"id": "seg-2", "labelId": "06", "label": "热食加工与出餐", "start": 65.0, "end": 140.0},
{"id": "seg-3", "labelId": "14", "label": "收银扫码", "start": 140.0, "end": 180.0},
{"id": "seg-4", "labelId": "15", "label": "顾客交付与内部协作", "start": 180.0, "end": 200.0},
],
}
api_call("POST", f"/api/plugins/annotation-qa/tasks/{task_id}/revisions", ann_rev_2, user_header=annotator_auth)
print(" • 标注员补充标签 15 并提交 Revision 2 (返工状态: OutputDelta=0, 不重复计算产出)")
# 8.7 Quality reviewer claims and approves Revision 2
cur_item = api_call("GET", f"/api/human-items/{item_id}", user_header=quality_auth)
api_call("POST", f"/api/human-items/{item_id}/claim", {"version": get_item_version(cur_item)}, user_header=quality_auth)
qc_ann_approve = {
"stage": "quality",
"decision": "approved",
"videoQuality": "qualified",
"annotationAccuracy": 99.0,
"annotationCompleteness": 99.5,
"overallGrade": "A",
"comment": "补充标签准确完整,时序对齐精准,质检通过",
}
api_call("POST", f"/api/plugins/annotation-qa/tasks/{task_id}/reviews", qc_ann_approve, user_header=quality_auth)
print(" ✅ 标注质检通过,任务成功晋升为 accepted 状态!")
# 9. Verify Workload Summary & Rework Isolation
print("\n[Step 9/11] 校验供应商工时、产出与防重复计费指标...")
summary = api_call("GET", f"/api/plugins/annotation-qa/workload-summary?projectId={project_id}", user_header=admin_auth)
annotator_workload = next((item for item in summary.get("items", []) if item.get("userId") == annotator_user["id"]), None)
if annotator_workload:
print(" • 泛宇标注员工时与产出审计结果:")
print(f" -> 净计费产出 (OutputCount): {annotator_workload.get('outputCount')} (必须严格等于 1)")
print(f" -> 提交版本总数 (SubmittedCount): {annotator_workload.get('submittedCount')}")
print(f" -> 打回返工次数 (ReworkCount): {annotator_workload.get('reworkCount')} (准确记录返工)")
assert annotator_workload.get("outputCount") == 1, "Rework output duplicated!"
assert annotator_workload.get("reworkCount") >= 1, "Rework count missing!"
print(" ✅ 供应商工时统计与防重复计费审计 100% 通过!")
# 10. Delivery Package Acceptance
print("\n[Step 10/11] 阶段三:执行【交付包验收】(验收专家 A+ 评级与签署通过)...")
dpkg_id = f"dpkg-fanyu-10w-{task_id}"
acc_batch_id = f"hbatch-acc-{dpkg_id}"
acc_item_id = f"hitem-acc-{dpkg_id}"
# Insert acceptance human batch, item & delivery package into MongoDB
acc_batch_doc = {
"_id": acc_batch_id,
"id": acc_batch_id,
"organizationId": "org_default",
"pluginId": "delivery-acceptance",
"projectId": project_id,
"workflowId": workflow_id,
"runId": run_id,
"gateNodeId": "node-acceptance-gate",
"status": "active",
"stages": [
{"id": "acceptance", "label": "最终交付验收", "kind": "acceptance", "requiredRole": "acceptance_reviewer"},
],
"releasePolicy": "all_accepted",
"totalItemCount": 1,
"readyItemCount": 1,
"acceptedItemCount": 0,
"rejectedItemCount": 0,
"createdAt": now_iso(),
"updatedAt": now_iso(),
}
acc_item_doc = {
"_id": acc_item_id,
"id": acc_item_id,
"organizationId": "org_default",
"pluginId": "delivery-acceptance",
"batchId": acc_batch_id,
"projectId": project_id,
"workflowId": workflow_id,
"runId": run_id,
"gateNodeId": "node-acceptance-gate",
"workgroupId": workgroup_id,
"vendorId": "vendor-fanyu-a",
"subject": {"itemKey": f"{dpkg_id}.tar.gz", "mediaType": "application/json"},
"status": "ready",
"currentStageId": "acceptance",
"reviewRound": 1,
"version": 1,
"createdAt": now_iso(),
"updatedAt": now_iso(),
}
dpkg_doc = {
"_id": dpkg_id,
"id": dpkg_id,
"organizationId": "org_default",
"projectId": project_id,
"workflowId": workflow_id,
"parentRunId": run_id,
"childRunId": f"{run_id}-delivery",
"releasePartitionId": "part-1",
"manifestDigest": "sha256:fanyu10wmanifest1234567890",
"itemIds": [item_id],
"workgroupId": workgroup_id,
"vendorId": "vendor-fanyu-a",
"status": "pending_acceptance",
"acceptanceHumanItemId": acc_item_id,
"artifactIds": ["art-face-blur", "art-convenience-json", "art-rrd"],
"version": 1,
"createdAt": now_iso(),
"updatedAt": now_iso(),
}
db["humanBatches"].replace_one({"_id": acc_batch_id}, acc_batch_doc, upsert=True)
db["humanItems"].replace_one({"_id": acc_item_id}, acc_item_doc, upsert=True)
db["deliveryPackages"].replace_one({"_id": dpkg_id}, dpkg_doc, upsert=True)
# Acceptance expert claims acceptance human item & delivery package
cur_acc_item = api_call("GET", f"/api/human-items/{acc_item_id}", user_header=acceptance_auth)
api_call("POST", f"/api/human-items/{acc_item_id}/claim", {"version": get_item_version(cur_acc_item)}, user_header=acceptance_auth)
cur_dpkg = api_call("GET", f"/api/delivery-packages/{dpkg_id}", user_header=acceptance_auth)
api_call("POST", f"/api/delivery-packages/{dpkg_id}/claim", {"version": cur_dpkg["version"]}, user_header=acceptance_auth)
print(" • 验收专家 (李工) 认领交付包与验收卡片")
# Acceptance expert decision
cur_dpkg = api_call("GET", f"/api/delivery-packages/{dpkg_id}", user_header=acceptance_auth)
decision = {
"version": cur_dpkg["version"],
"decision": "approved",
"videoQuality": "qualified",
"annotationAccuracy": 99.5,
"annotationCompleteness": 100.0,
"overallGrade": "A+",
"comment": "泛宇 10w 小时首批便利店第一视角数据已通过全流程验收,准予交付并触发下游持续训练",
}
api_call("POST", f"/api/delivery-packages/{dpkg_id}/decisions", decision, user_header=acceptance_auth, expected_status=(200, 201))
print(" ✅ 验收专家评级 A+ 签署交付通过!")
# 11. Trigger Downstream Continuous Training
print("\n[Step 11/11] 阶段四:触发下游持续训练算子 (Qwen3-VL 22 类动作分类持续训练)...")
from backend.operators.builtin.qwen_label_crop_classification.operator import run as run_qwen_crop
train_dir = pathlib.Path("./build/continuous_training_output")
train_dir.mkdir(parents=True, exist_ok=True)
inputs = [
{
"name": "delivery_package",
"data": {
"sourceVideo": "bos://testxyz/fanyu_test/Orbbec_Ego_AZER76400GG_19700101_001_D.mp4",
"segments": [
{"labelId": "05", "label": "蒸点陈列与出餐", "start": 0.0, "end": 65.0},
{"labelId": "06", "label": "热食加工与出餐", "start": 65.0, "end": 140.0},
{"labelId": "14", "label": "收银扫码", "start": 140.0, "end": 180.0},
{"labelId": "15", "label": "顾客交付与内部协作", "start": 180.0, "end": 200.0},
],
},
}
]
artifacts = run_qwen_crop(
run_id=run_id,
node_id="node-train",
output_dir=train_dir,
node_data={"hyperparameters": {"TASK_MODE": "continuous_train", "MAX_CLIP_SECONDS": 200.0}},
inputs=inputs,
helpers={},
)
print(f" ✅ 持续训练算子执行完成!已在 {train_dir} 生成 {len(artifacts)} 个模型与评测产物:")
for art in artifacts:
print(f" • {art.get('name')}: {art.get('uri')}")
print("\n" + "=" * 80)
print("🎉🎉🎉 泛宇便利店第一视角 10w 小时人在环全流程在 POC 环境真实执行 100% 成功!")
print("=" * 80)
if __name__ == "__main__":
main()