[cherry-pick][MOT]set mot infer frame rate (PaddlePaddle#3876)

* add frame_rate for mot infer video

* add doc of frame_rate for mot video infer, test=document_fix

configs/mot/README.md

@@ -262,7 +262,7 @@ Inference a vidoe on single GPU with following command:
 
 ```bash
 # inference on video and save a video
-CUDA_VISIBLE_DEVICES=0 python tools/infer_mot.py -c configs/mot/fairmot/fairmot_dla34_30e_1088x608.yml -o weights=https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608.pdparams --video_file={your video name}.mp4  --save_videos
+CUDA_VISIBLE_DEVICES=0 python tools/infer_mot.py -c configs/mot/fairmot/fairmot_dla34_30e_1088x608.yml -o weights=https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608.pdparams --video_file={your video name}.mp4 --frame_rate=20 --save_videos
 ```
 
 Inference a image folder on single GPU with following command:
@@ -273,7 +273,7 @@ CUDA_VISIBLE_DEVICES=0 python tools/infer_mot.py -c configs/mot/fairmot/fairmot_
 ```
 
 **Notes:**
- Please make sure that [ffmpeg](https://ffmpeg.org/ffmpeg.html) is installed first, on Linux(Ubuntu) platform you can directly install it by the following command:`apt-get update && apt-get install -y ffmpeg`.
+ Please make sure that [ffmpeg](https://ffmpeg.org/ffmpeg.html) is installed first, on Linux(Ubuntu) platform you can directly install it by the following command:`apt-get update && apt-get install -y ffmpeg`. `--frame_rate` means the frame rate of the video and the frames extracted per second. It can be set by yourself, default value is -1 indicating the video frame rate read by OpenCV.
 
 
 ### 4. Export model

configs/mot/README_cn.md

@@ -260,7 +260,7 @@ EvalMOTDataset:
 
 ```bash
 # 预测一个视频
-CUDA_VISIBLE_DEVICES=0 python tools/infer_mot.py -c configs/mot/fairmot/fairmot_dla34_30e_1088x608.yml -o weights=https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608.pdparams --video_file={your video name}.mp4  --save_videos
+CUDA_VISIBLE_DEVICES=0 python tools/infer_mot.py -c configs/mot/fairmot/fairmot_dla34_30e_1088x608.yml -o weights=https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608.pdparams --video_file={your video name}.mp4 --frame_rate=20 --save_videos
 ```
 
 使用单个GPU通过如下命令预测一个图片文件夹，并保存为视频
@@ -271,7 +271,7 @@ CUDA_VISIBLE_DEVICES=0 python tools/infer_mot.py -c configs/mot/fairmot/fairmot_
 ```
 
 **注意:**
- 请先确保已经安装了[ffmpeg](https://ffmpeg.org/ffmpeg.html), Linux(Ubuntu)平台可以直接用以下命令安装：`apt-get update && apt-get install -y ffmpeg`。
+ 请先确保已经安装了[ffmpeg](https://ffmpeg.org/ffmpeg.html), Linux(Ubuntu)平台可以直接用以下命令安装：`apt-get update && apt-get install -y ffmpeg`。`--frame_rate`表示视频的帧率，表示每秒抽取多少帧，可以自行设置，默认为-1表示会使用OpenCV读取的视频帧率。
 
 ### 4. 导出预测模型
 

ppdet/data/source/mot.py

@@ -234,6 +234,7 @@ class MOTImageFolder(DetDataset):
     Load MOT dataset with MOT format from image folder or video .
     Args:
         video_file (str): path of the video file, default ''.
+        frame_rate (int): frame rate of the video, use cv2 VideoCapture if not set.
         dataset_dir (str): root directory for dataset.
         keep_ori_im (bool): whether to keep original image, default False. 
             Set True when used during MOT model inference while saving
@@ -242,6 +243,7 @@ class MOTImageFolder(DetDataset):
 
     def __init__(self,
                  video_file=None,
+                 frame_rate=-1,
                  dataset_dir=None,
                  data_root=None,
                  image_dir=None,
@@ -255,25 +257,29 @@ def __init__(self,
         self.keep_ori_im = keep_ori_im
         self._imid2path = {}
         self.roidbs = None
-        self.frame_rate = 30
+        self.frame_rate = frame_rate
 
     def check_or_download_dataset(self):
         return
 
     def parse_dataset(self, ):
         if not self.roidbs:
             if self.video_file is None:
+                self.frame_rate = 30  # set as default if infer image folder
                 self.roidbs = self._load_images()
             else:
                 self.roidbs = self._load_video_images()
 
     def _load_video_images(self):
-        cap = cv2.VideoCapture(self.video_file)
-        self.frame_rate = int(cap.get(cv2.CAP_PROP_FPS))
+        if self.frame_rate == -1:
+            # if frame_rate is not set for video, use cv2.VideoCapture
+            cap = cv2.VideoCapture(self.video_file)
+            self.frame_rate = int(cap.get(cv2.CAP_PROP_FPS))
 
         extension = self.video_file.split('.')[-1]
         output_path = self.video_file.replace('.{}'.format(extension), '')
-        frames_path = video2frames(self.video_file, output_path)
+        frames_path = video2frames(self.video_file, output_path,
+                                   self.frame_rate)
         self.video_frames = sorted(
             glob.glob(os.path.join(frames_path, '*.png')))
 
@@ -334,8 +340,10 @@ def set_images(self, images):
         self.image_dir = images
         self.roidbs = self._load_images()
 
-    def set_video(self, video_file):
+    def set_video(self, video_file, frame_rate):
+        # update video_file and frame_rate by command line of tools/infer_mot.py
         self.video_file = video_file
+        self.frame_rate = frame_rate
         assert os.path.isfile(self.video_file) and _is_valid_video(self.video_file), \
                 "wrong or unsupported file format: {}".format(self.video_file)
         self.roidbs = self._load_video_images()
@@ -345,7 +353,7 @@ def _is_valid_video(f, extensions=('.mp4', '.avi', '.mov', '.rmvb', 'flv')):
     return f.lower().endswith(extensions)
 
 
-def video2frames(video_path, outpath, **kargs):
+def video2frames(video_path, outpath, frame_rate, **kargs):
     def _dict2str(kargs):
         cmd_str = ''
         for k, v in kargs.items():
@@ -363,7 +371,9 @@ def _dict2str(kargs):
     outformat = os.path.join(out_full_path, '%08d.png')
 
     cmd = ffmpeg
-    cmd = ffmpeg + [' -i ', video_path, ' -start_number ', ' 0 ', outformat]
+    cmd = ffmpeg + [
+        ' -i ', video_path, ' -r ', str(frame_rate), ' -f image2 ', outformat
+    ]
     cmd = ''.join(cmd) + _dict2str(kargs)
 
     try:

ppdet/engine/tracker.py

@@ -374,6 +374,7 @@ def get_infer_images(self, infer_dir):
 
     def mot_predict(self,
                     video_file,
+                    frame_rate,
                     image_dir,
                     output_dir,
                     data_type='mot',
@@ -401,7 +402,7 @@ def mot_predict(self,
         # run tracking        
         if video_file:
             seq = video_file.split('/')[-1].split('.')[0]
-            self.dataset.set_video(video_file)
+            self.dataset.set_video(video_file, frame_rate)
             logger.info('Starting tracking video {}'.format(video_file))
         elif image_dir:
             seq = image_dir.split('/')[-1].split('.')[0]
@@ -420,7 +421,8 @@ def mot_predict(self,
 
         dataloader = create('TestMOTReader')(self.dataset, 0)
         result_filename = os.path.join(result_root, '{}.txt'.format(seq))
-        frame_rate = self.dataset.frame_rate
+        if frame_rate == -1:
+            frame_rate = self.dataset.frame_rate
 
         with paddle.no_grad():
             if model_type in ['JDE', 'FairMOT']:

tools/infer_mot.py

@@ -43,6 +43,11 @@ def parse_args():
     parser = ArgsParser()
     parser.add_argument(
         '--video_file', type=str, default=None, help='Video name for tracking.')
+    parser.add_argument(
+        '--frame_rate',
+        type=int,
+        default=-1,
+        help='Video frame rate for tracking.')
     parser.add_argument(
         "--image_dir",
         type=str,
@@ -95,6 +100,7 @@ def run(FLAGS, cfg):
     # inference
     tracker.mot_predict(
         video_file=FLAGS.video_file,
+        frame_rate=FLAGS.frame_rate,
         image_dir=FLAGS.image_dir,
         data_type=cfg.metric.lower(),
         model_type=cfg.architecture,