🧪 Require WJZH video timing

This commit is contained in:
eust-w 2026-06-16 21:29:08 +08:00
parent 953341240b
commit f30294b167
7 changed files with 222 additions and 26 deletions

View File

@ -71,7 +71,7 @@ Drobodata is a workflow orchestration platform for embodied AI and vision pipeli
- `npm run reconcile:data -- <api-base-url>`: remove legacy demo data, reset settings, and bootstrap built-in operators through the real API; honors `DROBOTICFLOW_API_TOKEN` and optional session login env vars for protected deployments
- `npm run acceptance:object-storage`: browser-driven object-storage acceptance against a configured live storage connection; requires `OBJECT_STORAGE_CONNECTION_ID`, `OBJECT_STORAGE_CONNECTION_NAME`, or `OBJECT_STORAGE_BUCKET`
- `npm run acceptance:xplatform`: X-platform operator-chain acceptance against an explicit real dataset; requires `XPLATFORM_SOURCE_PATH`, explicit `XPLATFORM_RULES_JSON`, all `XPLATFORM_MIN_*` counts, quality/slicing thresholds, and a storage connection selector when the source or upload path is `bos://...`
- `python3 scripts/wjzh_operator_regression.py --preflight-only`: preflight the production-style WJZH real-data operator regression against a live API; requires explicit `SOURCE_PATH`, `UPLOAD_ROOT_PREFIX`, storage connection selector, target video filenames, route/filter values, real Robotics dataset paths, incremental seed metadata, quality thresholds, and `SAMPLING_RULES_JSON`
- `python3 scripts/wjzh_operator_regression.py --preflight-only`: preflight the production-style WJZH real-data operator regression against a live API; requires explicit `SOURCE_PATH`, `UPLOAD_ROOT_PREFIX`, storage connection selector, target video filenames, route/filter values, real Robotics dataset paths, incremental seed metadata, quality thresholds, video timing, and `SAMPLING_RULES_JSON`
Builtin operator metadata is declared once in `internal/api/builtin_operators_catalog.json`. The API embeds and validates that catalog before seeding `/api/operators`, while `scripts/generate_operator_docs.py` reads the same file for runtime handler hints. Runtime deployment fields such as the active operator-runner image are injected at startup or reconciliation time.

View File

@ -132,11 +132,21 @@ INCREMENTAL_VIDEO_SUBDIR=<video-subdir> \
QUALITY_MIN_CLARITY=<threshold> \
QUALITY_MAX_SHAKE=<threshold> \
QUALITY_MAX_HAND_OCCLUSION=<threshold> \
HUGE_WINDOW_SECONDS=<seconds> \
MEDIUM_WINDOW_SECONDS=<seconds> \
MEDIUM_SEGMENT_DURATION_SECONDS=<seconds> \
HUGE_SEGMENT_DURATION_SECONDS=<seconds> \
BATCH_WINDOW_SECONDS=<seconds> \
BATCH_QUALITY_WINDOW_SECONDS=<seconds> \
BATCH_SEGMENT_DURATION_SECONDS=<seconds> \
INCREMENTAL_WINDOW_SECONDS=<seconds> \
INCREMENTAL_SEGMENT_DURATION_SECONDS=<seconds> \
ANNOTATION_FRAME_INTERVAL_SECONDS=<seconds> \
SAMPLING_RULES_JSON='<sampling-rules-json-array>' \
python3 scripts/wjzh_operator_regression.py --preflight-only
```
The WJZH regression script does not provide built-in quality or sampling thresholds; `QUALITY_MIN_CLARITY`, `QUALITY_MAX_SHAKE`, `QUALITY_MAX_HAND_OCCLUSION`, and `SAMPLING_RULES_JSON` must match the target dataset before the run is accepted as evidence. See the full runbook in [`docs/wjzh-operator-regression.md`](wjzh-operator-regression.md).
The WJZH regression script does not provide built-in quality thresholds, video timing, or sampling thresholds; `QUALITY_MIN_CLARITY`, `QUALITY_MAX_SHAKE`, `QUALITY_MAX_HAND_OCCLUSION`, the `*_SECONDS` timing values, and `SAMPLING_RULES_JSON` must match the target dataset before the run is accepted as evidence. See the full runbook in [`docs/wjzh-operator-regression.md`](wjzh-operator-regression.md).
Run browser object-storage acceptance against a live storage connection:

View File

@ -10,7 +10,7 @@ This note tracks the production-facing safeguards that must stay true for Drobod
- Explicit shell operators are disabled by default through `OPERATOR_ALLOW_SHELL_COMMANDS=false`; allowlist and denylist regexes are platform-controlled and cannot be overridden by workflow hyperparams.
- Environment-specific values such as `K8S_ENDPOINT`, registry username, and workspace BOS bucket have no private repository defaults; provide them from local deploy secrets or CI secrets.
- Data reconciliation uses real API calls and supports protected deployments through `DROBOTICFLOW_API_TOKEN` or session login credentials plus CSRF headers. Public session-auth environments must use an organization admin account for reconcile because it updates settings and management resources.
- Python live-acceptance and WJZH regression scripts use the shared `DroboticflowClient` helper for API token headers, session-login cookies, CSRF propagation, JSON calls, and binary object-storage calls. WJZH quality thresholds are caller-supplied through `QUALITY_MIN_CLARITY`, `QUALITY_MAX_SHAKE`, and `QUALITY_MAX_HAND_OCCLUSION`; sampling thresholds are caller-supplied through `SAMPLING_RULES_JSON`. The scripts do not carry built-in audit rules. New Python acceptance scripts should reuse that helper instead of carrying their own auth/request stack.
- Python live-acceptance and WJZH regression scripts use the shared `DroboticflowClient` helper for API token headers, session-login cookies, CSRF propagation, JSON calls, and binary object-storage calls. WJZH quality thresholds are caller-supplied through `QUALITY_MIN_CLARITY`, `QUALITY_MAX_SHAKE`, and `QUALITY_MAX_HAND_OCCLUSION`; window, slicing, and annotation timing are caller-supplied through the documented `*_SECONDS` values; sampling thresholds are caller-supplied through `SAMPLING_RULES_JSON`. The scripts do not carry built-in audit rules. New Python acceptance scripts should reuse that helper instead of carrying their own auth/request stack.
- Node/Playwright acceptance scripts use the shared `createAcceptanceAPIClient` helper for bearer token headers, Playwright context CSRF propagation, JSON calls, retries, and optional session login. Functional and operator-preview checks also pass `DROBOTICFLOW_API_TOKEN` through browser contexts so protected UI requests exercise the same token-authenticated path as API setup calls.
## Kubernetes Jobs

View File

@ -21,6 +21,16 @@ INCREMENTAL_VIDEO_SUBDIR=<video-subdir> \
QUALITY_MIN_CLARITY=<threshold> \
QUALITY_MAX_SHAKE=<threshold> \
QUALITY_MAX_HAND_OCCLUSION=<threshold> \
HUGE_WINDOW_SECONDS=<seconds> \
MEDIUM_WINDOW_SECONDS=<seconds> \
MEDIUM_SEGMENT_DURATION_SECONDS=<seconds> \
HUGE_SEGMENT_DURATION_SECONDS=<seconds> \
BATCH_WINDOW_SECONDS=<seconds> \
BATCH_QUALITY_WINDOW_SECONDS=<seconds> \
BATCH_SEGMENT_DURATION_SECONDS=<seconds> \
INCREMENTAL_WINDOW_SECONDS=<seconds> \
INCREMENTAL_SEGMENT_DURATION_SECONDS=<seconds> \
ANNOTATION_FRAME_INTERVAL_SECONDS=<seconds> \
SAMPLING_RULES_JSON='[{"id":"clarity-gate","field":"quality.clarity","operator":"gte","value":0.3,"samplePercent":100}]' \
python3 scripts/wjzh_operator_regression.py --preflight-only
```
@ -55,12 +65,14 @@ SOURCE_PATH=bos://<bucket>/<input-prefix>
UPLOAD_ROOT_PREFIX=bos://<bucket>/<output-prefix>
```
脚本不再内置任何真实 bucket、路径、连接名、视频文件名、设备号、日期范围、Robotics 数据路径、增量目录元数据、质检阈值或抽检规则。执行者必须按目标环境传入 `SOURCE_PATH`、`UPLOAD_ROOT_PREFIX`、`STORAGE_CONNECTION_ID` 或 `STORAGE_CONNECTION_NAME`、`MEDIUM_VIDEO`、`HUGE_VIDEO`、`ROUTE_DEVICE_ID`、`FILTER_DATE_START`、`FILTER_DATE_END`、`LEROBOT_SOURCE_PATH`、`ROSBAG_SOURCE_PATH`、`INCREMENTAL_PLACE`、`INCREMENTAL_CAMERA_MODEL`、`INCREMENTAL_VIDEO_SUBDIR`、`QUALITY_MIN_CLARITY`、`QUALITY_MAX_SHAKE`、`QUALITY_MAX_HAND_OCCLUSION`、`SAMPLING_RULES_JSON`。`AUDIT_VIDEO` 可单独指定;未指定时复用 `MEDIUM_VIDEO`。受保护部署还需要 `DROBOTICFLOW_API_TOKEN`,或 `DROBOTICFLOW_SMOKE_EMAIL` / `DROBOTICFLOW_SMOKE_PASSWORD` 组织管理员登录凭据。
脚本不再内置任何真实 bucket、路径、连接名、视频文件名、设备号、日期范围、Robotics 数据路径、增量目录元数据、质检阈值、视频处理时长或抽检规则。执行者必须按目标环境传入 `SOURCE_PATH`、`UPLOAD_ROOT_PREFIX`、`STORAGE_CONNECTION_ID` 或 `STORAGE_CONNECTION_NAME`、`MEDIUM_VIDEO`、`HUGE_VIDEO`、`ROUTE_DEVICE_ID`、`FILTER_DATE_START`、`FILTER_DATE_END`、`LEROBOT_SOURCE_PATH`、`ROSBAG_SOURCE_PATH`、`INCREMENTAL_PLACE`、`INCREMENTAL_CAMERA_MODEL`、`INCREMENTAL_VIDEO_SUBDIR`、`QUALITY_MIN_CLARITY`、`QUALITY_MAX_SHAKE`、`QUALITY_MAX_HAND_OCCLUSION`、`HUGE_WINDOW_SECONDS`、`MEDIUM_WINDOW_SECONDS`、`MEDIUM_SEGMENT_DURATION_SECONDS`、`HUGE_SEGMENT_DURATION_SECONDS`、`BATCH_WINDOW_SECONDS`、`BATCH_QUALITY_WINDOW_SECONDS`、`BATCH_SEGMENT_DURATION_SECONDS`、`INCREMENTAL_WINDOW_SECONDS`、`INCREMENTAL_SEGMENT_DURATION_SECONDS`、`ANNOTATION_FRAME_INTERVAL_SECONDS`、`SAMPLING_RULES_JSON`。`AUDIT_VIDEO` 可单独指定;未指定时复用 `MEDIUM_VIDEO`。受保护部署还需要 `DROBOTICFLOW_API_TOKEN`,或 `DROBOTICFLOW_SMOKE_EMAIL` / `DROBOTICFLOW_SMOKE_PASSWORD` 组织管理员登录凭据。
`SAMPLING_RULES_JSON` 必须是非空 JSON 数组。每条规则必须包含 `id`、`field`、`operator`、`value`、`samplePercent`,且 `samplePercent` 取值范围是 `(0, 100]`。`operator` 支持 `eq`、`ne`、`gt`、`gte`、`lt`、`lte`、`contains`、`in`、`exists`、`missing`。这些规则会写入 `07-超大视频全链路`、`11-多目录多视频批量全链路`、`12-增量嗅探循环处理` 以及多项目入口的 `11B-人工抽检采样`,脚本不会替调用方选择默认阈值。
`QUALITY_MIN_CLARITY`、`QUALITY_MAX_SHAKE`、`QUALITY_MAX_HAND_OCCLUSION` 必须是非负数字,会写入所有 `视频质检` 节点以及带日期过滤的 `数据过滤` 节点。调用方需要根据目标真实数据集选择阈值;脚本不会使用 `0.0/1.0/1.0` 这类宽松默认值作为验收标准。
窗口、切片和标注 timing 配置必须显式提供。`HUGE_WINDOW_SECONDS`、`MEDIUM_WINDOW_SECONDS`、`MEDIUM_SEGMENT_DURATION_SECONDS`、`HUGE_SEGMENT_DURATION_SECONDS`、`BATCH_WINDOW_SECONDS`、`BATCH_QUALITY_WINDOW_SECONDS`、`BATCH_SEGMENT_DURATION_SECONDS`、`INCREMENTAL_WINDOW_SECONDS`、`INCREMENTAL_SEGMENT_DURATION_SECONDS` 必须是正整数;`ANNOTATION_FRAME_INTERVAL_SECONDS` 必须是大于 0 的数字。这些值会写入 `Shell 命令执行` 的窗口采样命令、`视频切片` 的 `SEGMENT_DURATION_SECONDS` 和 `视频标注` 的 `FRAME_INTERVAL_SECONDS`。
## 生产前置
必须满足下面 4 项,否则脚本会在 preflight 或 runtime smoke 阶段直接失败:
@ -101,6 +113,16 @@ INCREMENTAL_VIDEO_SUBDIR=<video-subdir> \
QUALITY_MIN_CLARITY=<threshold> \
QUALITY_MAX_SHAKE=<threshold> \
QUALITY_MAX_HAND_OCCLUSION=<threshold> \
HUGE_WINDOW_SECONDS=<seconds> \
MEDIUM_WINDOW_SECONDS=<seconds> \
MEDIUM_SEGMENT_DURATION_SECONDS=<seconds> \
HUGE_SEGMENT_DURATION_SECONDS=<seconds> \
BATCH_WINDOW_SECONDS=<seconds> \
BATCH_QUALITY_WINDOW_SECONDS=<seconds> \
BATCH_SEGMENT_DURATION_SECONDS=<seconds> \
INCREMENTAL_WINDOW_SECONDS=<seconds> \
INCREMENTAL_SEGMENT_DURATION_SECONDS=<seconds> \
ANNOTATION_FRAME_INTERVAL_SECONDS=<seconds> \
SAMPLING_RULES_JSON='<sampling-rules-json-array>' \
python3 scripts/wjzh_operator_regression.py --preflight-only
```
@ -125,6 +147,16 @@ INCREMENTAL_VIDEO_SUBDIR=<video-subdir> \
QUALITY_MIN_CLARITY=<threshold> \
QUALITY_MAX_SHAKE=<threshold> \
QUALITY_MAX_HAND_OCCLUSION=<threshold> \
HUGE_WINDOW_SECONDS=<seconds> \
MEDIUM_WINDOW_SECONDS=<seconds> \
MEDIUM_SEGMENT_DURATION_SECONDS=<seconds> \
HUGE_SEGMENT_DURATION_SECONDS=<seconds> \
BATCH_WINDOW_SECONDS=<seconds> \
BATCH_QUALITY_WINDOW_SECONDS=<seconds> \
BATCH_SEGMENT_DURATION_SECONDS=<seconds> \
INCREMENTAL_WINDOW_SECONDS=<seconds> \
INCREMENTAL_SEGMENT_DURATION_SECONDS=<seconds> \
ANNOTATION_FRAME_INTERVAL_SECONDS=<seconds> \
SAMPLING_RULES_JSON='<sampling-rules-json-array>' \
python3 scripts/wjzh_operator_regression.py --delete-regression-projects
```
@ -161,6 +193,16 @@ INCREMENTAL_VIDEO_SUBDIR=<video-subdir> \
QUALITY_MIN_CLARITY=<threshold> \
QUALITY_MAX_SHAKE=<threshold> \
QUALITY_MAX_HAND_OCCLUSION=<threshold> \
HUGE_WINDOW_SECONDS=<seconds> \
MEDIUM_WINDOW_SECONDS=<seconds> \
MEDIUM_SEGMENT_DURATION_SECONDS=<seconds> \
HUGE_SEGMENT_DURATION_SECONDS=<seconds> \
BATCH_WINDOW_SECONDS=<seconds> \
BATCH_QUALITY_WINDOW_SECONDS=<seconds> \
BATCH_SEGMENT_DURATION_SECONDS=<seconds> \
INCREMENTAL_WINDOW_SECONDS=<seconds> \
INCREMENTAL_SEGMENT_DURATION_SECONDS=<seconds> \
ANNOTATION_FRAME_INTERVAL_SECONDS=<seconds> \
SAMPLING_RULES_JSON='<sampling-rules-json-array>' \
python3 scripts/wjzh_operator_regression.py --preflight-only
```
@ -185,6 +227,16 @@ INCREMENTAL_VIDEO_SUBDIR=<video-subdir> \
QUALITY_MIN_CLARITY=<threshold> \
QUALITY_MAX_SHAKE=<threshold> \
QUALITY_MAX_HAND_OCCLUSION=<threshold> \
HUGE_WINDOW_SECONDS=<seconds> \
MEDIUM_WINDOW_SECONDS=<seconds> \
MEDIUM_SEGMENT_DURATION_SECONDS=<seconds> \
HUGE_SEGMENT_DURATION_SECONDS=<seconds> \
BATCH_WINDOW_SECONDS=<seconds> \
BATCH_QUALITY_WINDOW_SECONDS=<seconds> \
BATCH_SEGMENT_DURATION_SECONDS=<seconds> \
INCREMENTAL_WINDOW_SECONDS=<seconds> \
INCREMENTAL_SEGMENT_DURATION_SECONDS=<seconds> \
ANNOTATION_FRAME_INTERVAL_SECONDS=<seconds> \
SAMPLING_RULES_JSON='<sampling-rules-json-array>' \
python3 scripts/wjzh_operator_regression.py --preflight-only
```
@ -216,6 +268,16 @@ INCREMENTAL_VIDEO_SUBDIR=<video-subdir> \
QUALITY_MIN_CLARITY=<threshold> \
QUALITY_MAX_SHAKE=<threshold> \
QUALITY_MAX_HAND_OCCLUSION=<threshold> \
HUGE_WINDOW_SECONDS=<seconds> \
MEDIUM_WINDOW_SECONDS=<seconds> \
MEDIUM_SEGMENT_DURATION_SECONDS=<seconds> \
HUGE_SEGMENT_DURATION_SECONDS=<seconds> \
BATCH_WINDOW_SECONDS=<seconds> \
BATCH_QUALITY_WINDOW_SECONDS=<seconds> \
BATCH_SEGMENT_DURATION_SECONDS=<seconds> \
INCREMENTAL_WINDOW_SECONDS=<seconds> \
INCREMENTAL_SEGMENT_DURATION_SECONDS=<seconds> \
ANNOTATION_FRAME_INTERVAL_SECONDS=<seconds> \
SAMPLING_RULES_JSON='<sampling-rules-json-array>' \
python3 scripts/wjzh_operator_regression.py --preflight-only
```
@ -244,6 +306,16 @@ INCREMENTAL_VIDEO_SUBDIR=<video-subdir> \
QUALITY_MIN_CLARITY=<threshold> \
QUALITY_MAX_SHAKE=<threshold> \
QUALITY_MAX_HAND_OCCLUSION=<threshold> \
HUGE_WINDOW_SECONDS=<seconds> \
MEDIUM_WINDOW_SECONDS=<seconds> \
MEDIUM_SEGMENT_DURATION_SECONDS=<seconds> \
HUGE_SEGMENT_DURATION_SECONDS=<seconds> \
BATCH_WINDOW_SECONDS=<seconds> \
BATCH_QUALITY_WINDOW_SECONDS=<seconds> \
BATCH_SEGMENT_DURATION_SECONDS=<seconds> \
INCREMENTAL_WINDOW_SECONDS=<seconds> \
INCREMENTAL_SEGMENT_DURATION_SECONDS=<seconds> \
ANNOTATION_FRAME_INTERVAL_SECONDS=<seconds> \
SAMPLING_RULES_JSON='<sampling-rules-json-array>' \
python3 scripts/wjzh_operator_multi_project_regression.py --preflight-only
```

View File

@ -251,7 +251,7 @@ def create_manual_sampling_workflow(
"batch_pick",
660,
0,
{"COMMAND": base.cap_duration_command(45, max_items=4, max_duration=500)},
{"COMMAND": base.cap_duration_command(base.int_config(config.batch_window_seconds), max_items=4, max_duration=500)},
label="人工抽检前抽样",
),
base.build_node(
@ -266,14 +266,14 @@ def create_manual_sampling_workflow(
"slice",
1100,
-60,
{"SEGMENT_DURATION_SECONDS": "30"},
base.segment_hyperparams(config.batch_segment_duration_seconds),
),
base.build_node(
operators_by_name["视频标注"],
"annotate",
1320,
-60,
{"FRAME_INTERVAL_SECONDS": "1.0"},
base.annotation_hyperparams(config),
),
base.build_node(operators_by_name["成品打包"], "package", 1540, -60),
base.build_node(

View File

@ -24,7 +24,6 @@ from lib.droboticflow_acceptance import DroboticflowClient
DEFAULT_API_BASE_URL = "http://127.0.0.1:3001"
DEFAULT_PROJECT_PREFIX = "真实数据算子回归"
DEFAULT_HUGE_WINDOW_SECONDS = 120
CLIENT: DroboticflowClient | None = None
@ -58,7 +57,7 @@ class Config:
audit_video: str
medium_video: str
huge_video: str
huge_window_seconds: int
huge_window_seconds: str
route_device_id: str
filter_date_start: str
filter_date_end: str
@ -71,6 +70,15 @@ class Config:
quality_min_clarity: str
quality_max_shake: str
quality_max_hand_occlusion: str
medium_window_seconds: str
medium_segment_duration_seconds: str
huge_segment_duration_seconds: str
batch_window_seconds: str
batch_quality_window_seconds: str
batch_segment_duration_seconds: str
incremental_window_seconds: str
incremental_segment_duration_seconds: str
annotation_frame_interval_seconds: str
def env_default(name: str, fallback: str = "") -> str:
@ -145,6 +153,18 @@ def filter_hyperparams(config: Config) -> dict[str, str]:
}
def int_config(value: str) -> int:
return int(str(value).strip())
def segment_hyperparams(duration_seconds: str) -> dict[str, str]:
return {"SEGMENT_DURATION_SECONDS": str(duration_seconds).strip()}
def annotation_hyperparams(config: Config) -> dict[str, str]:
return {"FRAME_INTERVAL_SECONDS": config.annotation_frame_interval_seconds}
def validate_config(config: Config) -> None:
issues: list[str] = []
parsed_filter_start: time.struct_time | None = None
@ -223,6 +243,37 @@ def validate_config(config: Config) -> None:
continue
if parsed_value < 0:
issues.append(f"{label} must be greater than or equal to 0")
for label, value in (
("HUGE_WINDOW_SECONDS", config.huge_window_seconds),
("MEDIUM_WINDOW_SECONDS", config.medium_window_seconds),
("MEDIUM_SEGMENT_DURATION_SECONDS", config.medium_segment_duration_seconds),
("HUGE_SEGMENT_DURATION_SECONDS", config.huge_segment_duration_seconds),
("BATCH_WINDOW_SECONDS", config.batch_window_seconds),
("BATCH_QUALITY_WINDOW_SECONDS", config.batch_quality_window_seconds),
("BATCH_SEGMENT_DURATION_SECONDS", config.batch_segment_duration_seconds),
("INCREMENTAL_WINDOW_SECONDS", config.incremental_window_seconds),
("INCREMENTAL_SEGMENT_DURATION_SECONDS", config.incremental_segment_duration_seconds),
):
if not value:
issues.append(f"{label} is required")
continue
try:
parsed_value = int(value)
except ValueError:
issues.append(f"{label} must be an integer")
continue
if parsed_value <= 0:
issues.append(f"{label} must be greater than 0")
if not config.annotation_frame_interval_seconds:
issues.append("ANNOTATION_FRAME_INTERVAL_SECONDS is required")
else:
try:
annotation_interval = float(config.annotation_frame_interval_seconds)
except ValueError:
issues.append("ANNOTATION_FRAME_INTERVAL_SECONDS must be numeric")
else:
if annotation_interval <= 0:
issues.append("ANNOTATION_FRAME_INTERVAL_SECONDS must be greater than 0")
if issues:
raise RegressionError(
"invalid regression configuration:\n"
@ -269,7 +320,7 @@ def parse_args() -> Config:
parser.add_argument("--audit-video", default=env_default("AUDIT_VIDEO"))
parser.add_argument("--medium-video", default=env_default("MEDIUM_VIDEO"))
parser.add_argument("--huge-video", default=env_default("HUGE_VIDEO"))
parser.add_argument("--huge-window-seconds", type=int, default=int(env_default("HUGE_WINDOW_SECONDS", str(DEFAULT_HUGE_WINDOW_SECONDS)) or DEFAULT_HUGE_WINDOW_SECONDS))
parser.add_argument("--huge-window-seconds", default=env_default("HUGE_WINDOW_SECONDS"))
parser.add_argument("--route-device-id", default=env_default("ROUTE_DEVICE_ID"))
parser.add_argument("--filter-date-start", default=env_default("FILTER_DATE_START"))
parser.add_argument("--filter-date-end", default=env_default("FILTER_DATE_END"))
@ -282,6 +333,15 @@ def parse_args() -> Config:
parser.add_argument("--quality-min-clarity", default=env_default("QUALITY_MIN_CLARITY"))
parser.add_argument("--quality-max-shake", default=env_default("QUALITY_MAX_SHAKE"))
parser.add_argument("--quality-max-hand-occlusion", default=env_default("QUALITY_MAX_HAND_OCCLUSION"))
parser.add_argument("--medium-window-seconds", default=env_default("MEDIUM_WINDOW_SECONDS"))
parser.add_argument("--medium-segment-duration-seconds", default=env_default("MEDIUM_SEGMENT_DURATION_SECONDS"))
parser.add_argument("--huge-segment-duration-seconds", default=env_default("HUGE_SEGMENT_DURATION_SECONDS"))
parser.add_argument("--batch-window-seconds", default=env_default("BATCH_WINDOW_SECONDS"))
parser.add_argument("--batch-quality-window-seconds", default=env_default("BATCH_QUALITY_WINDOW_SECONDS"))
parser.add_argument("--batch-segment-duration-seconds", default=env_default("BATCH_SEGMENT_DURATION_SECONDS"))
parser.add_argument("--incremental-window-seconds", default=env_default("INCREMENTAL_WINDOW_SECONDS"))
parser.add_argument("--incremental-segment-duration-seconds", default=env_default("INCREMENTAL_SEGMENT_DURATION_SECONDS"))
parser.add_argument("--annotation-frame-interval-seconds", default=env_default("ANNOTATION_FRAME_INTERVAL_SECONDS"))
args = parser.parse_args()
return Config(
api_base_url=args.api_base_url.rstrip("/"),
@ -308,7 +368,7 @@ def parse_args() -> Config:
audit_video=args.audit_video or args.medium_video,
medium_video=args.medium_video,
huge_video=args.huge_video,
huge_window_seconds=max(args.huge_window_seconds, 1),
huge_window_seconds=args.huge_window_seconds,
route_device_id=args.route_device_id,
filter_date_start=args.filter_date_start,
filter_date_end=args.filter_date_end,
@ -321,6 +381,15 @@ def parse_args() -> Config:
quality_min_clarity=args.quality_min_clarity,
quality_max_shake=args.quality_max_shake,
quality_max_hand_occlusion=args.quality_max_hand_occlusion,
medium_window_seconds=args.medium_window_seconds,
medium_segment_duration_seconds=args.medium_segment_duration_seconds,
huge_segment_duration_seconds=args.huge_segment_duration_seconds,
batch_window_seconds=args.batch_window_seconds,
batch_quality_window_seconds=args.batch_quality_window_seconds,
batch_segment_duration_seconds=args.batch_segment_duration_seconds,
incremental_window_seconds=args.incremental_window_seconds,
incremental_segment_duration_seconds=args.incremental_segment_duration_seconds,
annotation_frame_interval_seconds=args.annotation_frame_interval_seconds,
)
@ -1255,8 +1324,15 @@ def create_regression_workflows(
build_node(validate_operator, "validate", 440, 0, {"REQUIRE_INTRINSICS": "false"}),
build_node(json_filter_operator, "pick_medium", 660, 0, {"FIELD_PATH": "filename", "OPERATOR": "eq", "EXPECTED": config.medium_video}),
build_node(operators_by_name["视频质检"], "quality", 880, -60, quality_hyperparams(config)),
build_node(shell_operator, "window_medium", 1100, -60, {"COMMAND": cap_duration_command(45, max_items=1, max_duration=120)}, label="中视频窗口采样"),
build_node(operators_by_name["视频切片"], "slice", 1320, -60, {"SEGMENT_DURATION_SECONDS": "45"}),
build_node(
shell_operator,
"window_medium",
1100,
-60,
{"COMMAND": cap_duration_command(int_config(config.medium_window_seconds), max_items=1, max_duration=int_config(config.huge_window_seconds))},
label="中视频窗口采样",
),
build_node(operators_by_name["视频切片"], "slice", 1320, -60, segment_hyperparams(config.medium_segment_duration_seconds)),
named_upload("quality_upload", 1540, -140, "06-medium-video/quality"),
named_upload("slice_upload", 1540, 20, "06-medium-video/slice"),
]
@ -1289,9 +1365,9 @@ def create_regression_workflows(
0,
filter_hyperparams(config),
),
build_node(shell_operator, "window", 1320, 0, {"COMMAND": cap_duration_command(config.huge_window_seconds)}, label="大文件窗口采样"),
build_node(operators_by_name["视频切片"], "slice", 1540, 0, {"SEGMENT_DURATION_SECONDS": "120"}),
build_node(operators_by_name["视频标注"], "annotate", 1760, 0, {"FRAME_INTERVAL_SECONDS": "1.0"}),
build_node(shell_operator, "window", 1320, 0, {"COMMAND": cap_duration_command(int_config(config.huge_window_seconds))}, label="大文件窗口采样"),
build_node(operators_by_name["视频切片"], "slice", 1540, 0, segment_hyperparams(config.huge_segment_duration_seconds)),
build_node(operators_by_name["视频标注"], "annotate", 1760, 0, annotation_hyperparams(config)),
build_node(operators_by_name["成品打包"], "package", 1980, 0),
build_node(
operators_by_name["抽检采样"],
@ -1338,10 +1414,10 @@ def create_regression_workflows(
build_node(index_operator, "index", 220, 0),
build_node(validate_operator, "validate", 440, 0, {"REQUIRE_INTRINSICS": "false"}),
build_node(json_filter_operator, "pick_batch", 660, 0, {"FIELD_PATH": "duration", "OPERATOR": "lte", "EXPECTED_JSON": "400"}),
build_node(shell_operator, "window_batch", 880, 0, {"COMMAND": cap_duration_command(45, max_items=4)}, label="批量窗口采样"),
build_node(shell_operator, "window_batch", 880, 0, {"COMMAND": cap_duration_command(int_config(config.batch_window_seconds), max_items=4)}, label="批量窗口采样"),
build_node(operators_by_name["视频质检"], "quality", 1100, 0, quality_hyperparams(config)),
build_node(operators_by_name["视频切片"], "slice", 1320, 0, {"SEGMENT_DURATION_SECONDS": "30"}),
build_node(operators_by_name["视频标注"], "annotate", 1540, 0, {"FRAME_INTERVAL_SECONDS": "1.0"}),
build_node(operators_by_name["视频切片"], "slice", 1320, 0, segment_hyperparams(config.batch_segment_duration_seconds)),
build_node(operators_by_name["视频标注"], "annotate", 1540, 0, annotation_hyperparams(config)),
build_node(operators_by_name["成品打包"], "package", 1760, 0),
named_upload("batch_package_upload", 1980, -60, "08-batch-windowed/package"),
named_upload("batch_annotation_upload", 1980, 60, "08-batch-windowed/annotate"),
@ -1384,7 +1460,14 @@ def create_regression_workflows(
scan_source(),
build_node(index_operator, "index", 220, 0),
build_node(validate_operator, "validate", 440, 0, {"REQUIRE_INTRINSICS": "false"}),
build_node(shell_operator, "batch_pick", 660, 0, {"COMMAND": cap_duration_command(90, max_items=8, max_duration=500)}, label="多目录多视频抽样"),
build_node(
shell_operator,
"batch_pick",
660,
0,
{"COMMAND": cap_duration_command(int_config(config.batch_quality_window_seconds), max_items=8, max_duration=500)},
label="多目录多视频抽样",
),
build_node(operators_by_name["视频质检"], "quality", 900, 0, quality_hyperparams(config)),
named_upload("batch_pick_upload", 1140, -60, "10-batch-quality/pick"),
named_upload("quality_upload", 1140, 60, "10-batch-quality/quality"),
@ -1408,11 +1491,11 @@ def create_regression_workflows(
scan_source(),
build_node(index_operator, "index", 220, 0),
build_node(validate_operator, "validate", 440, 0, {"REQUIRE_INTRINSICS": "false"}),
build_node(shell_operator, "batch_pick", 660, 0, {"COMMAND": cap_duration_command(45, max_items=6, max_duration=500)}, label="多目录全链路抽样"),
build_node(shell_operator, "batch_pick", 660, 0, {"COMMAND": cap_duration_command(int_config(config.batch_window_seconds), max_items=6, max_duration=500)}, label="多目录全链路抽样"),
build_node(operators_by_name["视频质检"], "quality", 900, -80, quality_hyperparams(config)),
build_node(operators_by_name["数据过滤"], "filter", 1120, -80, filter_hyperparams(config)),
build_node(operators_by_name["视频切片"], "slice", 1340, -80, {"SEGMENT_DURATION_SECONDS": "30"}),
build_node(operators_by_name["视频标注"], "annotate", 1560, -80, {"FRAME_INTERVAL_SECONDS": "1.0"}),
build_node(operators_by_name["视频切片"], "slice", 1340, -80, segment_hyperparams(config.batch_segment_duration_seconds)),
build_node(operators_by_name["视频标注"], "annotate", 1560, -80, annotation_hyperparams(config)),
build_node(operators_by_name["成品打包"], "package", 1780, -80),
build_node(
operators_by_name["抽检采样"],
@ -1458,9 +1541,16 @@ def create_regression_workflows(
build_node(validate_operator, "validate", 440, 0, {"REQUIRE_INTRINSICS": "false"}),
build_node(operators_by_name["视频质检"], "quality", 660, -80, quality_hyperparams(config)),
build_node(operators_by_name["数据过滤"], "filter", 880, -80, filter_hyperparams(config)),
build_node(shell_operator, "window", 1100, -80, {"COMMAND": cap_duration_command(30, max_items=8, max_duration=120)}, label="增量视频窗口抽样"),
build_node(operators_by_name["视频切片"], "slice", 1320, -80, {"SEGMENT_DURATION_SECONDS": "30"}),
build_node(operators_by_name["视频标注"], "annotate", 1540, -80, {"FRAME_INTERVAL_SECONDS": "1.0"}),
build_node(
shell_operator,
"window",
1100,
-80,
{"COMMAND": cap_duration_command(int_config(config.incremental_window_seconds), max_items=8, max_duration=int_config(config.huge_window_seconds))},
label="增量视频窗口抽样",
),
build_node(operators_by_name["视频切片"], "slice", 1320, -80, segment_hyperparams(config.incremental_segment_duration_seconds)),
build_node(operators_by_name["视频标注"], "annotate", 1540, -80, annotation_hyperparams(config)),
build_node(operators_by_name["成品打包"], "package", 1760, -80),
build_node(
operators_by_name["抽检采样"],

View File

@ -75,6 +75,15 @@ def valid_config(**overrides):
"quality_min_clarity": "0.35",
"quality_max_shake": "0.2",
"quality_max_hand_occlusion": "0.25",
"medium_window_seconds": "45",
"medium_segment_duration_seconds": "45",
"huge_segment_duration_seconds": "120",
"batch_window_seconds": "45",
"batch_quality_window_seconds": "90",
"batch_segment_duration_seconds": "30",
"incremental_window_seconds": "30",
"incremental_segment_duration_seconds": "30",
"annotation_frame_interval_seconds": "1.0",
}
values.update(overrides)
return BASE.Config(**values)
@ -107,6 +116,15 @@ class WJZHOperatorRegressionTest(unittest.TestCase):
with self.assertRaisesRegex(BASE.RegressionError, "QUALITY_MAX_SHAKE"):
BASE.validate_config(config)
def test_validate_config_requires_explicit_video_timing(self):
config = valid_config(batch_segment_duration_seconds="", annotation_frame_interval_seconds="0")
with self.assertRaisesRegex(BASE.RegressionError, "BATCH_SEGMENT_DURATION_SECONDS"):
BASE.validate_config(config)
with self.assertRaisesRegex(BASE.RegressionError, "ANNOTATION_FRAME_INTERVAL_SECONDS"):
BASE.validate_config(config)
def test_normalize_sampling_rules_json_validates_rule_shape(self):
with self.assertRaisesRegex(BASE.RegressionError, "samplePercent"):
BASE.normalize_sampling_rules_json(
@ -185,6 +203,12 @@ class WJZHOperatorRegressionTest(unittest.TestCase):
self.assertEqual(quality_params["MIN_CLARITY"], "0.35")
self.assertEqual(quality_params["MAX_SHAKE"], "0.2")
self.assertEqual(quality_params["MAX_HAND_OCCLUSION"], "0.25")
slice_node = next(node for node in created_payloads[0]["nodes"] if node["id"] == "slice")
slice_params = {item["key"]: item["value"] for item in slice_node["data"]["hyperparams"]}
self.assertEqual(slice_params["SEGMENT_DURATION_SECONDS"], "30")
annotate_node = next(node for node in created_payloads[0]["nodes"] if node["id"] == "annotate")
annotate_params = {item["key"]: item["value"] for item in annotate_node["data"]["hyperparams"]}
self.assertEqual(annotate_params["FRAME_INTERVAL_SECONDS"], "1.0")
if __name__ == "__main__":