[to #42322933] video detecor support output with timestamp

适配demoservice，增加视频时间戳输出，每一个结果对应一个时间戳
        Link: https://code.alibaba-inc.com/Ali-MaaS/MaaS-lib/codereview/10496879

modelscope/models/cv/realtime_object_detection/realtime_video_detector.py

@@ -16,6 +16,7 @@ from modelscope.models.builder import MODELS
 from modelscope.preprocessors import LoadImage
 from modelscope.utils.config import Config
 from modelscope.utils.constant import ModelFile, Tasks
+from .utils import timestamp_format
 from .yolox.data.data_augment import ValTransform
 from .yolox.exp import get_exp_by_name
 from .yolox.utils import postprocess
@@ -99,14 +100,17 @@ class RealtimeVideoDetector(TorchModel):
     def inference_video(self, v_path):
         outputs = []
         desc = 'Detecting video: {}'.format(v_path)
-        for frame, result in tqdm(
-                self.inference_video_iter(v_path), desc=desc):
+        for frame_idx, (frame, result) in enumerate(
+                tqdm(self.inference_video_iter(v_path), desc=desc)):
+            result = result + (timestamp_format(seconds=frame_idx
+                                                / self.fps), )
             outputs.append(result)
 
         return outputs
 
     def inference_video_iter(self, v_path):
         capture = cv2.VideoCapture(v_path)
+        self.fps = capture.get(cv2.CAP_PROP_FPS)
         while capture.isOpened():
             ret, frame = capture.read()
             if not ret:

modelscope/models/cv/realtime_object_detection/utils.py

@@ -0,0 +1,9 @@
+# Copyright (c) Alibaba, Inc. and its affiliates.
+import math
+
+
+def timestamp_format(seconds):
+    m, s = divmod(seconds, 60)
+    h, m = divmod(m, 60)
+    time = '%02d:%02d:%06.3f' % (h, m, s)
+    return time

modelscope/pipelines/cv/realtime_video_object_detection_pipeline.py

@@ -45,15 +45,17 @@ class RealtimeVideoObjectDetectionPipeline(Pipeline):
                     **kwargs) -> str:
         forward_output = input['forward_output']
 
-        scores, boxes, labels = [], [], []
+        scores, boxes, labels, timestamps = [], [], [], []
         for result in forward_output:
-            box, score, label = result
+            box, score, label, timestamp = result
             scores.append(score)
             boxes.append(box)
             labels.append(label)
+            timestamps.append(timestamp)
 
         return {
             OutputKeys.BOXES: boxes,
             OutputKeys.SCORES: scores,
             OutputKeys.LABELS: labels,
+            OutputKeys.TIMESTAMPS: timestamps,
         }