[MOT] Upgrade pptracking deploy (#5358)

* add bytetrack yolov3 engine and deploy baseline

* fix bytetrack yolov3 configs

* fix cfgs tools

* fix tracker pred_dets

* fit pptracking jde sde infer

* rename cfgs, fix mot_sde_infer

* fix mot sde single class infer

* remove byte cfgs

* fix mtmct

* clean code

deploy/pptracking/python/README.md

@@ -58,11 +58,12 @@ python deploy/pptracking/python/mot_jde_infer.py --model_dir=output_inference/fa
 ### 2.1 导出预测模型
 Step 1：导出检测模型
 ```bash
-# 导出JDE YOLOv3行人检测模型
-CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c configs/mot/deepsort/detector/jde_yolov3_darknet53_30e_1088x608_mix.yml -o weights=https://paddledet.bj.bcebos.com/models/mot/deepsort/jde_yolov3_darknet53_30e_1088x608_mix.pdparams
-# 或导出PPYOLOv2行人检测模型
+# 导出PPYOLOv2行人检测模型
 CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c configs/mot/deepsort/detector/ppyolov2_r50vd_dcn_365e_640x640_mot17half.yml -o weights=https://paddledet.bj.bcebos.com/models/mot/deepsort/ppyolov2_r50vd_dcn_365e_640x640_mot17half.pdparams
+# 或导出PPYOLOe行人检测模型
+CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c configs/mot/deepsort/detector/ppyoloe_crn_l_36e_640x640_mot17half.yml -o weights=https://paddledet.bj.bcebos.com/models/mot/deepsort/ppyoloe_crn_l_36e_640x640_mot17half.pdparams
 ```
+
 Step 2：导出行人ReID模型
 ```bash
 # 导出PCB Pyramid ReID模型
@@ -76,11 +77,10 @@ CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c configs/mot/deepsort/reid
 # 下载行人跟踪demo视频：
 wget https://bj.bcebos.com/v1/paddledet/data/mot/demo/mot17_demo.mp4
 
-# 用导出的JDE YOLOv3行人检测模型和PCB Pyramid ReID模型
-python deploy/pptracking/python/mot_sde_infer.py --model_dir=output_inference/jde_yolov3_darknet53_30e_1088x608_mix/ --reid_model_dir=output_inference/deepsort_pcb_pyramid_r101/ --video_file=mot17_demo.mp4 --device=GPU --threshold=0.5 --save_mot_txts --save_images
-
-# 或用导出的PPYOLOv2行人检测模型和PPLCNet ReID模型
-python deploy/pptracking/python/mot_sde_infer.py --model_dir=output_inference/ppyolov2_r50vd_dcn_365e_640x640_mot17half/ --reid_model_dir=output_inference/deepsort_pplcnet/ --video_file=mot17_demo.mp4 --device=GPU --threshold=0.5 --scaled=True --save_mot_txts --save_images
+# 用导出的PPYOLOv2行人检测模型和PPLCNet ReID模型
+python deploy/pptracking/python/mot_sde_infer.py --model_dir=output_inference/ppyolov2_r50vd_dcn_365e_640x640_mot17half/ --reid_model_dir=output_inference/deepsort_pplcnet/ --video_file=mot17_demo.mp4 --device=GPU --threshold=0.5 --save_mot_txts --save_images
+# 或用导出的PPYOLOe行人检测模型和PPLCNet ReID模型
+python deploy/pptracking/python/mot_sde_infer.py --model_dir=output_inference/ppyoloe_crn_l_36e_640x640_mot17half/ --reid_model_dir=output_inference/deepsort_pplcnet/ --video_file=mot17_demo.mp4 --device=GPU --threshold=0.5 --save_mot_txts --save_images
 ```
 
 ### 2.3 用导出的模型基于Python去预测车辆跟踪
@@ -97,17 +97,16 @@ wget https://paddledet.bj.bcebos.com/models/mot/deepsort/deepsort_pplcnet_vehicl
 tar -xvf deepsort_pplcnet_vehicle.tar
 
 # 用导出的PicoDet车辆检测模型和PPLCNet车辆ReID模型
-python deploy/pptracking/python/mot_sde_infer.py --model_dir=picodet_l_640_aic21mtmct_vehicle/ --reid_model_dir=deepsort_pplcnet_vehicle/ --device=GPU --scaled=True --threshold=0.5 --video_file={your video}.mp4 --save_mot_txts --save_images
+python deploy/pptracking/python/mot_sde_infer.py --model_dir=picodet_l_640_aic21mtmct_vehicle/ --reid_model_dir=deepsort_pplcnet_vehicle/ --device=GPU --threshold=0.5 --video_file={your video}.mp4 --save_mot_txts --save_images
 
 # 用导出的PP-YOLOv2车辆检测模型和PPLCNet车辆ReID模型
-python deploy/pptracking/python/mot_sde_infer.py --model_dir=ppyolov2_r50vd_dcn_365e_aic21mtmct_vehicle/ --reid_model_dir=deepsort_pplcnet_vehicle/ --device=GPU --scaled=True --threshold=0.5 --video_file={your video}.mp4 --save_mot_txts --save_images
+python deploy/pptracking/python/mot_sde_infer.py --model_dir=ppyolov2_r50vd_dcn_365e_aic21mtmct_vehicle/ --reid_model_dir=deepsort_pplcnet_vehicle/ --device=GPU --threshold=0.5 --video_file={your video}.mp4 --save_mot_txts --save_images
 ```
 
 **注意:**
  - 跟踪模型是对视频进行预测，不支持单张图的预测，默认保存跟踪结果可视化后的视频，可添加`--save_mot_txts`(对每个视频保存一个txt)或`--save_images`表示保存跟踪结果可视化图片。
  - 跟踪结果txt文件每行信息是`frame,id,x1,y1,w,h,score,-1,-1,-1`。
  - `--threshold`表示结果可视化的置信度阈值，默认为0.5，低于该阈值的结果会被过滤掉，为了可视化效果更佳，可根据实际情况自行修改。
- - `--scaled`表示在模型输出结果的坐标是否已经是缩放回原图的，如果使用的检测模型是JDE的YOLOv3则为False，如果使用通用检测模型则为True。
  - DeepSORT算法不支持多类别跟踪，只支持单类别跟踪，且ReID模型最好是与检测模型同一类别的物体训练过的，比如行人跟踪最好使用行人ReID模型，车辆跟踪最好使用车辆ReID模型。
 
 
@@ -135,94 +134,22 @@ wget https://paddledet.bj.bcebos.com/data/mot/demo/mtmct-demo.tar
 tar -xvf mtmct-demo.tar
 
 # 用导出的PicoDet车辆检测模型和PPLCNet车辆ReID模型
-python deploy/pptracking/python/mot_sde_infer.py --model_dir=picodet_l_640_aic21mtmct_vehicle/ --reid_model_dir=deepsort_pplcnet_vehicle/ --mtmct_dir=mtmct-demo --mtmct_cfg=mtmct_cfg --device=GPU --scaled=True --threshold=0.5 --save_mot_txts --save_images
+python deploy/pptracking/python/mot_sde_infer.py --model_dir=picodet_l_640_aic21mtmct_vehicle/ --reid_model_dir=deepsort_pplcnet_vehicle/ --mtmct_dir=mtmct-demo --mtmct_cfg=mtmct_cfg.yml --device=GPU --threshold=0.5 --save_mot_txts --save_images
 
 # 用导出的PP-YOLOv2车辆检测模型和PPLCNet车辆ReID模型
-python deploy/pptracking/python/mot_sde_infer.py --model_dir=ppyolov2_r50vd_dcn_365e_aic21mtmct_vehicle/ --reid_model_dir=deepsort_pplcnet_vehicle/ --mtmct_dir=mtmct-demo --mtmct_cfg=mtmct_cfg --device=GPU --scaled=True --threshold=0.5 --save_mot_txts --save_images
+python deploy/pptracking/python/mot_sde_infer.py --model_dir=ppyolov2_r50vd_dcn_365e_aic21mtmct_vehicle/ --reid_model_dir=deepsort_pplcnet_vehicle/ --mtmct_dir=mtmct-demo --mtmct_cfg=mtmct_cfg.yml --device=GPU --threshold=0.5 --save_mot_txts --save_images
 
 ```
 **注意:**
  - 跟踪模型是对视频进行预测，不支持单张图的预测，默认保存跟踪结果可视化后的视频，可添加`--save_mot_txts`(对每个视频保存一个txt)，或`--save_images`表示保存跟踪结果可视化图片。
  - 跨镜头跟踪结果txt文件每行信息是`camera_id,frame,id,x1,y1,w,h,-1,-1`。
  - `--threshold`表示结果可视化的置信度阈值，默认为0.5，低于该阈值的结果会被过滤掉，为了可视化效果更佳，可根据实际情况自行修改。
- - `--scaled`表示在模型输出结果的坐标是否已经是缩放回原图的，如果使用的检测模型是JDE的YOLOv3则为False，如果使用通用检测模型则为True。
  - DeepSORT算法不支持多类别跟踪，只支持单类别跟踪，且ReID模型最好是与检测模型同一类别的物体训练过的，比如行人跟踪最好使用行人ReID模型，车辆跟踪最好使用车辆ReID模型。
  - `--mtmct_dir`是MTMCT预测的某个场景的文件夹名字，里面包含该场景不同摄像头拍摄视频的图片文件夹，其数量至少为两个。
  - `--mtmct_cfg`是MTMCT预测的某个场景的配置文件，里面包含该一些trick操作的开关和该场景摄像头相关设置的文件路径，用户可以自行更改相关路径以及设置某些操作是否启用。
 
 
-## 4. API调用方式:
-
-### 4.1 FairMOT模型API调用
-```
-import mot_jde_infer
-
-# 1.model config and weights
-model_dir = 'fairmot_hrnetv2_w18_dlafpn_30e_576x320/'
-
-# 2.inference data
-video_file = 'test.mp4'
-image_dir = None
-
-# 3.other settings
-device = 'CPU' # device should be CPU, GPU or XPU
-threshold = 0.3
-output_dir = 'output'
-
-# mot predict
-mot_jde_infer.predict_naive(model_dir, video_file, image_dir, device, threshold, output_dir)
-```
-**注意:**
- - 以上代码必须进入目录`PaddleDetection/deploy/pptracking/python`下执行。
- - 支持对视频和图片文件夹进行预测，不支持单张图的预测，`video_file`或`image_dir`不能同时为None，推荐使用`video_file`，而`image_dir`需直接存放命名顺序规范的图片。
- - 默认会保存跟踪结果可视化后的图片和视频，以及跟踪结果txt文件，默认不会进行轨迹可视化和流量统计。
-
-
-### 4.2 DeepSORT模型API调用
-```
-import mot_sde_infer
-
-# 1.model config and weights
-model_dir = 'ppyolov2_r50vd_dcn_365e_aic21mtmct_vehicle/'
-reid_model_dir = 'deepsort_pplcnet_vehicle/'
-
-# 2.inference data
-video_file = 'test.mp4'
-image_dir = None
-
-# 3.other settings
-scaled = True # set False only when use JDE YOLOv3
-device = 'CPU' # device should be CPU, GPU or XPU
-threshold = 0.3
-output_dir = 'output'
-
-# 4. MTMCT settings, default None
-mtmct_dir = None
-mtmct_cfg = None
-
-# mot predict
-mot_sde_infer.predict_naive(model_dir,
-                            reid_model_dir,
-                            video_file,
-                            image_dir,
-                            mtmct_dir,
-                            mtmct_cfg,
-                            scaled,
-                            device,
-                            threshold,
-                            output_dir)
-```
-**注意:**
- - 以上代码必须进入目录`PaddleDetection/deploy/pptracking/python`下执行。
- - 支持对视频和图片文件夹进行预测，不支持单张图的预测，`video_file`或`image_dir`或`--mtmct_dir`不能同时为None，推荐使用`video_file`，而`image_dir`需直接存放命名顺序规范的图片，`--mtmct_dir`不为None表示是进行的MTMCT跨镜头跟踪任务。
- - 默认会保存跟踪结果可视化后的图片和视频，以及跟踪结果txt文件，默认不会进行轨迹可视化和流量统计。
- - `--scaled`表示在模型输出结果的坐标是否已经是缩放回原图的，如果使用的检测模型是JDE的YOLOv3则为False，如果使用通用检测模型则为True。
- - `--mtmct_dir`是MTMCT预测的某个场景的文件夹名字，里面包含该场景不同摄像头拍摄视频的图片文件夹，其数量至少为两个。
- - `--mtmct_cfg`是MTMCT预测的某个场景的配置文件，里面包含该一些trick操作的开关和该场景摄像头相关设置的文件路径，用户可以自行更改相关路径以及设置某些操作是否启用。
- - 开启MTMCT预测必须将`video_file`和`image_dir`同时设置为None，且`--mtmct_dir`和`--mtmct_cfg`都必须不为None。
-
-
-## 5. 参数说明:
+## 4. 参数说明:
 
 | 参数 | 是否必须|含义 |
 |-------|-------|----------|

deploy/pptracking/python/benchmark_utils.py

@@ -89,6 +89,8 @@ class PaddleInferBenchmark(object):
 
         self.preprocess_time_s = perf_info.get('preprocess_time_s', 0)
         self.postprocess_time_s = perf_info.get('postprocess_time_s', 0)
+        self.with_tracker = True if 'tracking_time_s' in perf_info else False
+        self.tracking_time_s = perf_info.get('tracking_time_s', 0)
         self.total_time_s = perf_info.get('total_time_s', 0)
 
         self.inference_time_s_90 = perf_info.get("inference_time_s_90", "")
@@ -235,8 +237,18 @@ class PaddleInferBenchmark(object):
         )
         self.logger.info(
             f"{identifier} total time spent(s): {self.total_time_s}")
+
+        if self.with_tracker:
+            self.logger.info(
+                f"{identifier} preprocess_time(ms): {round(self.preprocess_time_s*1000, 1)}, "
+                f"inference_time(ms): {round(self.inference_time_s*1000, 1)}, "
+                f"postprocess_time(ms): {round(self.postprocess_time_s*1000, 1)}, "
+                f"tracking_time(ms): {round(self.tracking_time_s*1000, 1)}")
+        else:
             self.logger.info(
-            f"{identifier} preprocess_time(ms): {round(self.preprocess_time_s*1000, 1)}, inference_time(ms): {round(self.inference_time_s*1000, 1)}, postprocess_time(ms): {round(self.postprocess_time_s*1000, 1)}"
+                f"{identifier} preprocess_time(ms): {round(self.preprocess_time_s*1000, 1)}, "
+                f"inference_time(ms): {round(self.inference_time_s*1000, 1)}, "
+                f"postprocess_time(ms): {round(self.postprocess_time_s*1000, 1)}"
             )
         if self.inference_time_s_90:
             self.looger.info(

deploy/pptracking/python/mot/tracker/deepsort_tracker.py

@@ -96,7 +96,8 @@ class DeepSORTTracker(object):
         """
         pred_cls_ids = pred_dets[:, 0:1]
         pred_scores = pred_dets[:, 1:2]
-        pred_tlwhs = pred_dets[:, 2:6]
+        pred_xyxys = pred_dets[:, 2:6]
+        pred_tlwhs = np.concatenate((pred_xyxys[:, 0:2], pred_xyxys[:, 2:4] - pred_xyxys[:, 0:2] + 1), axis=1)
 
         detections = [
             Detection(tlwh, score, feat, cls_id)

deploy/pptracking/python/mot/utils.py

@@ -182,8 +182,7 @@ def clip_box(xyxy, ori_image_shape):
 def get_crops(xyxy, ori_img, w, h):
     crops = []
     xyxy = xyxy.astype(np.int64)
-    ori_img = ori_img.numpy()
-    ori_img = np.squeeze(ori_img, axis=0).transpose(1, 0, 2)  # [h,w,3]->[w,h,3]
+    ori_img = ori_img.transpose(1, 0, 2) # [h,w,3]->[w,h,3]
     for i, bbox in enumerate(xyxy):
         crop = ori_img[bbox[0]:bbox[2], bbox[1]:bbox[3], :]
         crops.append(crop)
@@ -198,7 +197,10 @@ def preprocess_reid(imgs,
                     std=[0.229, 0.224, 0.225]):
     im_batch = []
     for img in imgs:
+        try:
             img = cv2.resize(img, (w, h))
+        except:
+            embed()
         img = img[:, :, ::-1].astype('float32').transpose((2, 0, 1)) / 255
         img_mean = np.array(mean).reshape((3, 1, 1))
         img_std = np.array(std).reshape((3, 1, 1))

deploy/pptracking/python/tracker_config.yml

+# config of tracker for MOT SDE Detector, use ByteTracker as default.
+# The tracker of MOT JDE Detector is exported together with the model.
+# Here 'min_box_area' and 'vertical_ratio' are set for pedestrian, you can modify for other objects tracking.
+tracker:
+  use_byte: true
+  conf_thres: 0.6
+  low_conf_thres: 0.1
+  match_thres: 0.9
+  min_box_area: 100
+  vertical_ratio: 1.6

deploy/pptracking/python/utils.py

@@ -66,6 +66,11 @@ def argsparser():
         default='cpu',
         help="Choose the device you want to run, it can be: CPU/GPU/XPU, default is CPU."
     )
+    parser.add_argument(
+        "--use_gpu",
+        type=ast.literal_eval,
+        default=False,
+        help="Deprecated, please use `--device`.")
     parser.add_argument(
         "--run_benchmark",
         type=ast.literal_eval,
@@ -104,12 +109,18 @@ def argsparser():
         '--save_mot_txts',
         action='store_true',
         help='Save tracking results (txt).')
+    parser.add_argument(
+        '--save_mot_txt_per_img',
+        action='store_true',
+        help='Save tracking results (txt) for each image.')
     parser.add_argument(
         '--scaled',
         type=bool,
         default=False,
         help="Whether coords after detector outputs are scaled, False in JDE YOLOv3 "
         "True in general detector.")
+    parser.add_argument(
+        "--tracker_config", type=str, default=None, help=("tracker donfig"))
     parser.add_argument(
         "--reid_model_dir",
         type=str,
@@ -122,20 +133,10 @@ def argsparser():
         default=50,
         help="max batch_size for reid model inference.")
     parser.add_argument(
-        "--do_entrance_counting",
-        action='store_true',
-        help="Whether counting the numbers of identifiers entering "
-        "or getting out from the entrance. Note that only support one-class"
-        "counting, multi-class counting is coming soon.")
-    parser.add_argument(
-        "--secs_interval",
-        type=int,
-        default=2,
-        help="The seconds interval to count after tracking")
-    parser.add_argument(
-        "--draw_center_traj",
-        action='store_true',
-        help="Whether drawing the trajectory of center")
+        '--use_dark',
+        type=ast.literal_eval,
+        default=True,
+        help='whether to use darkpose to get better keypoint position predict ')
     parser.add_argument(
         "--mtmct_dir",
         type=str,
@@ -146,6 +147,7 @@ def argsparser():
     return parser
 
 
+
 class Times(object):
     def __init__(self):
         self.time = 0.
@@ -174,29 +176,36 @@ class Times(object):
 
 
 class Timer(Times):
-    def __init__(self):
+    def __init__(self, with_tracker=False):
         super(Timer, self).__init__()
+        self.with_tracker = with_tracker
         self.preprocess_time_s = Times()
         self.inference_time_s = Times()
         self.postprocess_time_s = Times()
+        self.tracking_time_s = Times()
         self.img_num = 0
 
     def info(self, average=False):
-        total_time = self.preprocess_time_s.value(
-        ) + self.inference_time_s.value() + self.postprocess_time_s.value()
+        pre_time = self.preprocess_time_s.value()
+        infer_time = self.inference_time_s.value()
+        post_time = self.postprocess_time_s.value()
+        track_time = self.tracking_time_s.value()
+
+        total_time = pre_time + infer_time + post_time
+        if self.with_tracker:
+            total_time = total_time + track_time
         total_time = round(total_time, 4)
         print("------------------ Inference Time Info ----------------------")
         print("total_time(ms): {}, img_num: {}".format(total_time * 1000,
                                                        self.img_num))
-        preprocess_time = round(
-            self.preprocess_time_s.value() / max(1, self.img_num),
-            4) if average else self.preprocess_time_s.value()
-        postprocess_time = round(
-            self.postprocess_time_s.value() / max(1, self.img_num),
-            4) if average else self.postprocess_time_s.value()
-        inference_time = round(self.inference_time_s.value() /
-                               max(1, self.img_num),
-                               4) if average else self.inference_time_s.value()
+        preprocess_time = round(pre_time / max(1, self.img_num),
+                                4) if average else pre_time
+        postprocess_time = round(post_time / max(1, self.img_num),
+                                 4) if average else post_time
+        inference_time = round(infer_time / max(1, self.img_num),
+                               4) if average else infer_time
+        tracking_time = round(track_time / max(1, self.img_num),
+                              4) if average else track_time
 
         average_latency = total_time / max(1, self.img_num)
         qps = 0
@@ -204,6 +213,12 @@ class Timer(Times):
             qps = 1 / average_latency
         print("average latency time(ms): {:.2f}, QPS: {:2f}".format(
             average_latency * 1000, qps))
+        if self.with_tracker:
+            print(
+                "preprocess_time(ms): {:.2f}, inference_time(ms): {:.2f}, postprocess_time(ms): {:.2f}, tracking_time(ms): {:.2f}".
+                format(preprocess_time * 1000, inference_time * 1000,
+                       postprocess_time * 1000, tracking_time * 1000))
+        else:
             print(
                 "preprocess_time(ms): {:.2f}, inference_time(ms): {:.2f}, postprocess_time(ms): {:.2f}".
                 format(preprocess_time * 1000, inference_time * 1000,
@@ -211,18 +226,23 @@ class Timer(Times):
 
     def report(self, average=False):
         dic = {}
-        dic['preprocess_time_s'] = round(
-            self.preprocess_time_s.value() / max(1, self.img_num),
-            4) if average else self.preprocess_time_s.value()
-        dic['postprocess_time_s'] = round(
-            self.postprocess_time_s.value() / max(1, self.img_num),
-            4) if average else self.postprocess_time_s.value()
-        dic['inference_time_s'] = round(
-            self.inference_time_s.value() / max(1, self.img_num),
-            4) if average else self.inference_time_s.value()
+        pre_time = self.preprocess_time_s.value()
+        infer_time = self.inference_time_s.value()
+        post_time = self.postprocess_time_s.value()
+        track_time = self.tracking_time_s.value()
+
+        dic['preprocess_time_s'] = round(pre_time / max(1, self.img_num),
+                                         4) if average else pre_time
+        dic['inference_time_s'] = round(infer_time / max(1, self.img_num),
+                                        4) if average else infer_time
+        dic['postprocess_time_s'] = round(post_time / max(1, self.img_num),
+                                          4) if average else post_time
         dic['img_num'] = self.img_num
-        total_time = self.preprocess_time_s.value(
-        ) + self.inference_time_s.value() + self.postprocess_time_s.value()
+        total_time = pre_time + infer_time + post_time
+        if self.with_tracker:
+            dic['tracking_time_s'] = round(track_time / max(1, self.img_num),
+                                        4) if average else track_time
+            total_time = total_time + track_time
         dic['total_time_s'] = round(total_time, 4)
         return dic
 

deploy/python/infer.py

@@ -31,7 +31,7 @@ sys.path.insert(0, parent_path)
 
 from benchmark_utils import PaddleInferBenchmark
 from picodet_postprocess import PicoDetPostProcess
-from preprocess import preprocess, Resize, NormalizeImage, Permute, PadStride, LetterBoxResize, WarpAffine
+from preprocess import preprocess, Resize, NormalizeImage, Permute, PadStride, LetterBoxResize, WarpAffine, decode_image
 from keypoint_preprocess import EvalAffine, TopDownEvalAffine, expand_crop
 from visualize import visualize_box_mask
 from utils import argsparser, Timer, get_current_memory_mb

deploy/python/mot_keypoint_unite_infer.py

@@ -25,7 +25,7 @@ from collections import defaultdict
 from mot_keypoint_unite_utils import argsparser
 from preprocess import decode_image
 from infer import print_arguments, get_test_images
-from mot_sde_infer import SDE_Detector, MOT_SDE_SUPPORT_MODELS
+from mot_sde_infer import SDE_Detector
 from mot_jde_infer import JDE_Detector, MOT_JDE_SUPPORT_MODELS
 from keypoint_infer import KeyPointDetector, KEYPOINT_SUPPORT_MODELS
 from det_keypoint_unite_infer import predict_with_given_det

deploy/python/mot_sde_infer.py

@@ -34,13 +34,6 @@ from pptracking.python.mot import JDETracker
 from pptracking.python.mot.utils import MOTTimer, write_mot_results
 from pptracking.python.visualize import plot_tracking, plot_tracking_dict
 
-# Global dictionary
-MOT_SDE_SUPPORT_MODELS = {
-    'DeepSORT',
-    'ByteTrack',
-    'YOLO',
-}
-
 
 class SDE_Detector(Detector):
     """
@@ -287,7 +280,6 @@ def main():
     with open(deploy_file) as f:
         yml_conf = yaml.safe_load(f)
     arch = yml_conf['arch']
-    assert arch in MOT_SDE_SUPPORT_MODELS, '{} is not supported.'.format(arch)
     detector = SDE_Detector(
         FLAGS.model_dir,
         FLAGS.tracker_config,

deploy/python/utils.py

@@ -228,11 +228,11 @@ class Timer(Times):
                                         4) if average else infer_time
         dic['postprocess_time_s'] = round(post_time / max(1, self.img_num),
                                           4) if average else post_time
-        dic['tracking_time_s'] = round(post_time / max(1, self.img_num),
-                                       4) if average else track_time
         dic['img_num'] = self.img_num
         total_time = pre_time + infer_time + post_time
         if self.with_tracker:
+            dic['tracking_time_s'] = round(track_time / max(1, self.img_num),
+                                        4) if average else track_time
             total_time = total_time + track_time
         dic['total_time_s'] = round(total_time, 4)
         return dic

deploy/python/visualize.py

-# coding: utf-8
-# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.

ppdet/engine/export_utils.py

@@ -41,6 +41,7 @@ TRT_MIN_SUBGRAPH = {
     'HigherHRNet': 3,
     'HRNet': 3,
     'DeepSORT': 3,
+    'ByteTrack':10,
     'JDE': 10,
     'FairMOT': 5,
     'GFL': 16,
@@ -50,7 +51,7 @@ TRT_MIN_SUBGRAPH = {
 }
 
 KEYPOINT_ARCH = ['HigherHRNet', 'TopDownHRNet']
-MOT_ARCH = ['DeepSORT', 'JDE', 'FairMOT']
+MOT_ARCH = ['DeepSORT', 'JDE', 'FairMOT', 'ByteTrack']
 
 
 def _prune_input_spec(input_spec, program, targets):

ppdet/engine/tracker.py

@@ -29,6 +29,7 @@ from ppdet.core.workspace import create
 from ppdet.utils.checkpoint import load_weight, load_pretrain_weight
 from ppdet.modeling.mot.utils import Detection, get_crops, scale_coords, clip_box
 from ppdet.modeling.mot.utils import MOTTimer, load_det_results, write_mot_results, save_vis_results
+from ppdet.modeling.mot.tracker import JDETracker, DeepSORTTracker
 
 from ppdet.metrics import Metric, MOTMetric, KITTIMOTMetric
 from ppdet.metrics import MCMOTMetric
@@ -39,6 +40,11 @@ from .callbacks import Callback, ComposeCallback
 from ppdet.utils.logger import setup_logger
 logger = setup_logger(__name__)
 
+MOT_ARCH = ['DeepSORT', 'JDE', 'FairMOT', 'ByteTrack']
+MOT_ARCH_JDE = ['JDE', 'FairMOT']
+MOT_ARCH_SDE = ['DeepSORT', 'ByteTrack']
+MOT_DATA_TYPE = ['mot', 'mcmot', 'kitti']
+
 __all__ = ['Tracker']
 
 
@@ -109,11 +115,15 @@ class Tracker(object):
         load_weight(self.model, weights, self.optimizer)
 
     def load_weights_sde(self, det_weights, reid_weights):
-        if self.model.detector:
-            load_weight(self.model.detector, det_weights)
+        with_detector = self.model.detector is not None
+        with_reid = self.model.reid is not None
+
+        if with_detector:
+            load_weight(self.model.detector, det_weights, self.optimizer)
+            if with_reid:
                 load_weight(self.model.reid, reid_weights)
         else:
-            load_weight(self.model.reid, reid_weights, self.optimizer)
+            load_weight(self.model.reid, reid_weights)
 
     def _eval_seq_jde(self,
                       dataloader,
@@ -185,18 +195,21 @@ class Tracker(object):
         if save_dir:
             if not os.path.exists(save_dir): os.makedirs(save_dir)
         use_detector = False if not self.model.detector else True
+        use_reid = False if not self.model.reid else True
 
         timer = MOTTimer()
         results = defaultdict(list)
         frame_id = 0
         self.status['mode'] = 'track'
         self.model.eval()
+        if use_reid:
             self.model.reid.eval()
         if not use_detector:
             dets_list = load_det_results(det_file, len(dataloader))
             logger.info('Finish loading detection results file {}.'.format(
                 det_file))
 
+        tracker = self.model.tracker
         for step_id, data in enumerate(dataloader):
             self.status['step_id'] = step_id
             if frame_id % 40 == 0:
@@ -257,6 +270,8 @@ class Tracker(object):
                                                    scale_factor)
                     else:
                         pred_bboxes = outs['bbox'][:, 2:]
+                    pred_dets_old = np.concatenate(
+                        (pred_cls_ids, pred_scores, pred_bboxes), axis=1)
                 else:
                     logger.warning(
                         'Frame {} has not detected object, try to modify score threshold.'.
@@ -284,14 +299,10 @@ class Tracker(object):
 
             pred_cls_ids = pred_cls_ids[keep_idx[0]]
             pred_scores = pred_scores[keep_idx[0]]
-            pred_tlwhs = np.concatenate(
-                (pred_xyxys[:, 0:2],
-                 pred_xyxys[:, 2:4] - pred_xyxys[:, 0:2] + 1),
-                axis=1)
             pred_dets = np.concatenate(
-                (pred_cls_ids, pred_scores, pred_tlwhs), axis=1)
+                (pred_cls_ids, pred_scores, pred_xyxys), axis=1)
 
-            tracker = self.model.tracker
+            if use_reid:
                 crops = get_crops(
                     pred_xyxys,
                     ori_image,
@@ -301,11 +312,13 @@ class Tracker(object):
 
                 data.update({'crops': crops})
                 pred_embs = self.model(data).numpy()
+            else:
+                pred_embs = None
 
+            if isinstance(tracker, DeepSORTTracker):
+                online_tlwhs, online_scores, online_ids = [], [], []
                 tracker.predict()
                 online_targets = tracker.update(pred_dets, pred_embs)
-
-            online_tlwhs, online_scores, online_ids = [], [], []
                 for t in online_targets:
                     if not t.is_confirmed() or t.time_since_update > 1:
                         continue
@@ -328,6 +341,38 @@ class Tracker(object):
                 save_vis_results(data, frame_id, online_ids, online_tlwhs,
                                 online_scores, timer.average_time, show_image,
                                 save_dir, self.cfg.num_classes)
+
+            elif isinstance(tracker, JDETracker):
+                # trick hyperparams only used for MOTChallenge (MOT17, MOT20) Test-set
+                tracker.track_buffer, tracker.conf_thres = get_trick_hyperparams(
+                    seq_name, tracker.track_buffer, tracker.conf_thres)
+
+                online_targets_dict = tracker.update(pred_dets_old, pred_embs)
+                online_tlwhs = defaultdict(list)
+                online_scores = defaultdict(list)
+                online_ids = defaultdict(list)
+                for cls_id in range(self.cfg.num_classes):
+                    online_targets = online_targets_dict[cls_id]
+                    for t in online_targets:
+                        tlwh = t.tlwh
+                        tid = t.track_id
+                        tscore = t.score
+                        if tlwh[2] * tlwh[3] <= tracker.min_box_area: continue
+                        if tracker.vertical_ratio > 0 and tlwh[2] / tlwh[
+                                3] > tracker.vertical_ratio:
+                            continue
+                        online_tlwhs[cls_id].append(tlwh)
+                        online_ids[cls_id].append(tid)
+                        online_scores[cls_id].append(tscore)
+                    # save results
+                    results[cls_id].append(
+                        (frame_id + 1, online_tlwhs[cls_id], online_scores[cls_id],
+                        online_ids[cls_id]))
+                timer.toc()
+                save_vis_results(data, frame_id, online_ids, online_tlwhs,
+                                online_scores, timer.average_time, show_image,
+                                save_dir, self.cfg.num_classes)
+
             frame_id += 1
 
         return results, frame_id, timer.average_time, timer.calls
@@ -346,10 +391,10 @@ class Tracker(object):
         if not os.path.exists(output_dir): os.makedirs(output_dir)
         result_root = os.path.join(output_dir, 'mot_results')
         if not os.path.exists(result_root): os.makedirs(result_root)
-        assert data_type in ['mot', 'mcmot', 'kitti'], \
+        assert data_type in MOT_DATA_TYPE, \
             "data_type should be 'mot', 'mcmot' or 'kitti'"
-        assert model_type in ['JDE', 'DeepSORT', 'FairMOT'], \
-            "model_type should be 'JDE', 'DeepSORT' or 'FairMOT'"
+        assert model_type in MOT_ARCH, \
+            "model_type should be 'JDE', 'DeepSORT', 'FairMOT' or 'ByteTrack'"
 
         # run tracking
         n_frame = 0
@@ -380,13 +425,13 @@ class Tracker(object):
             result_filename = os.path.join(result_root, '{}.txt'.format(seq))
 
             with paddle.no_grad():
-                if model_type in ['JDE', 'FairMOT']:
+                if model_type in MOT_ARCH_JDE:
                     results, nf, ta, tc = self._eval_seq_jde(
                         dataloader,
                         save_dir=save_dir,
                         show_image=show_image,
                         frame_rate=frame_rate)
-                elif model_type in ['DeepSORT']:
+                elif model_type in MOT_ARCH_SDE:
                     results, nf, ta, tc = self._eval_seq_sde(
                         dataloader,
                         save_dir=save_dir,
@@ -472,10 +517,10 @@ class Tracker(object):
         if not os.path.exists(output_dir): os.makedirs(output_dir)
         result_root = os.path.join(output_dir, 'mot_results')
         if not os.path.exists(result_root): os.makedirs(result_root)
-        assert data_type in ['mot', 'mcmot', 'kitti'], \
+        assert data_type in MOT_DATA_TYPE, \
             "data_type should be 'mot', 'mcmot' or 'kitti'"
-        assert model_type in ['JDE', 'DeepSORT', 'FairMOT'], \
-            "model_type should be 'JDE', 'DeepSORT' or 'FairMOT'"
+        assert model_type in MOT_ARCH, \
+            "model_type should be 'JDE', 'DeepSORT', 'FairMOT' or 'ByteTrack'"
 
         # run tracking        
         if video_file:
@@ -505,14 +550,14 @@ class Tracker(object):
             frame_rate = self.dataset.frame_rate
 
         with paddle.no_grad():
-            if model_type in ['JDE', 'FairMOT']:
+            if model_type in MOT_ARCH_JDE:
                 results, nf, ta, tc = self._eval_seq_jde(
                     dataloader,
                     save_dir=save_dir,
                     show_image=show_image,
                     frame_rate=frame_rate,
                     draw_threshold=draw_threshold)
-            elif model_type in ['DeepSORT']:
+            elif model_type in MOT_ARCH_SDE:
                 results, nf, ta, tc = self._eval_seq_sde(
                     dataloader,
                     save_dir=save_dir,
@@ -536,3 +581,34 @@ class Tracker(object):
 
         write_mot_results(result_filename, results, data_type,
                           self.cfg.num_classes)
+
+def get_trick_hyperparams(video_name, ori_buffer, ori_thresh):
+    if video_name[:3] != 'MOT':
+        # only used for MOTChallenge (MOT17, MOT20) Test-set
+        return ori_buffer, ori_thresh
+
+    video_name = video_name[:8]
+    if 'MOT17-05' in video_name:
+        track_buffer = 14
+    elif 'MOT17-13' in video_name:
+        track_buffer = 25
+    else:
+        track_buffer = ori_buffer
+
+    if 'MOT17-01' in video_name:
+        track_thresh = 0.65
+    elif 'MOT17-06' in video_name:
+        track_thresh = 0.65
+    elif 'MOT17-12' in video_name:
+        track_thresh = 0.7
+    elif 'MOT17-14' in video_name:
+        track_thresh = 0.67
+    else:
+        track_thresh = ori_thresh
+
+    if 'MOT20-06' in video_name or 'MOT20-08' in video_name:
+        track_thresh = 0.3
+    else:
+        track_thresh = ori_thresh
+    
+    return track_buffer, ori_thresh

ppdet/engine/trainer.py

@@ -51,7 +51,7 @@ logger = setup_logger('ppdet.engine')
 
 __all__ = ['Trainer']
 
-MOT_ARCH = ['DeepSORT', 'JDE', 'FairMOT']
+MOT_ARCH = ['DeepSORT', 'JDE', 'FairMOT', 'ByteTrack']
 
 
 class Trainer(object):

ppdet/metrics/mcmot_metrics.py

@@ -308,7 +308,7 @@ class MCMOTEvaluator(object):
 
     def load_annotations(self):
         assert self.data_type == 'mcmot'
-        self.gt_filename = os.path.join(self.data_root, '../', '../',
+        self.gt_filename = os.path.join(self.data_root, '../',
                                         'sequences',
                                         '{}.txt'.format(self.seq_name))
         

ppdet/modeling/architectures/__init__.py

@@ -27,6 +27,7 @@ from . import detr
 from . import sparse_rcnn
 from . import tood
 from . import retinanet
+from . import bytetrack
 
 from .meta_arch import *
 from .faster_rcnn import *
@@ -51,3 +52,4 @@ from .detr import *
 from .sparse_rcnn import *
 from .tood import *
 from .retinanet import *
+from .bytetrack import *

ppdet/modeling/architectures/bytetrack.py

+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved. 
+#   
+# Licensed under the Apache License, Version 2.0 (the "License");   
+# you may not use this file except in compliance with the License.  
+# You may obtain a copy of the License at   
+#   
+#     http://www.apache.org/licenses/LICENSE-2.0    
+# 
+# Unless required by applicable law or agreed to in writing, software   
+# distributed under the License is distributed on an "AS IS" BASIS, 
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.  
+# See the License for the specific language governing permissions and   
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+from ppdet.core.workspace import register, create
+from .meta_arch import BaseArch
+
+__all__ = ['ByteTrack']
+
+
+@register
+class ByteTrack(BaseArch):
+    """
+    ByteTrack network, see https://arxiv.org/abs/
+
+    Args:
+        detector (object): detector model instance
+        reid (object): reid model instance
+        tracker (object): tracker instance
+    """
+    __category__ = 'architecture'
+
+    def __init__(self,
+                 detector='YOLOX',
+                 reid=None,
+                 tracker='JDETracker'):
+        super(ByteTrack, self).__init__()
+        self.detector = detector
+        self.reid = reid
+        self.tracker = tracker
+
+    @classmethod
+    def from_config(cls, cfg, *args, **kwargs):
+        detector = create(cfg['detector'])
+
+        if cfg['reid'] != 'None':
+            reid = create(cfg['reid'])
+        else:
+            reid = None
+
+        tracker = create(cfg['tracker'])
+
+        return {
+            "detector": detector,
+            "reid": reid,
+            "tracker": tracker,
+        }
+
+    def _forward(self):
+        det_outs = self.detector(self.inputs)
+
+        if self.training:
+            return det_outs
+        else:
+            if self.reid is not None:
+                assert 'crops' in self.inputs
+                crops = self.inputs['crops']
+                pred_embs = self.reid(crops)
+            else:
+                pred_embs = None
+            det_outs['embeddings'] = pred_embs
+            return det_outs
+
+    def get_loss(self):
+        return self._forward()
+
+    def get_pred(self):
+        return self._forward()
+

ppdet/modeling/mot/tracker/deepsort_tracker.py

@@ -102,7 +102,8 @@ class DeepSORTTracker(object):
         """
         pred_cls_ids = pred_dets[:, 0:1]
         pred_scores = pred_dets[:, 1:2]
-        pred_tlwhs = pred_dets[:, 2:6]
+        pred_xyxys = pred_dets[:, 2:6]
+        pred_tlwhs = np.concatenate((pred_xyxys[:, 0:2], pred_xyxys[:, 2:4] - pred_xyxys[:, 0:2] + 1), axis=1)
 
         detections = [
             Detection(tlwh, score, feat, cls_id)

tools/eval_mot.py

@@ -34,9 +34,6 @@ from ppdet.engine import Tracker
 from ppdet.utils.check import check_gpu, check_version, check_config
 from ppdet.utils.cli import ArgsParser
 
-from ppdet.utils.logger import setup_logger
-logger = setup_logger('eval')
-
 
 def parse_args():
     parser = ArgsParser()
@@ -83,11 +80,8 @@ def run(FLAGS, cfg):
     tracker = Tracker(cfg, mode='eval')
 
     # load weights
-    if cfg.architecture in ['DeepSORT']:
-        if cfg.det_weights != 'None':
+    if cfg.architecture in ['DeepSORT', 'ByteTrack']:
         tracker.load_weights_sde(cfg.det_weights, cfg.reid_weights)
-        else:
-            tracker.load_weights_sde(None, cfg.reid_weights)
     else:
         tracker.load_weights_jde(cfg.weights)
 

tools/export_model.py

@@ -28,7 +28,6 @@ import warnings
 warnings.filterwarnings('ignore')
 
 import paddle
-
 from ppdet.core.workspace import load_config, merge_config
 from ppdet.utils.check import check_gpu, check_version, check_config
 from ppdet.utils.cli import ArgsParser
@@ -65,11 +64,8 @@ def run(FLAGS, cfg):
     trainer = Trainer(cfg, mode='test')
 
     # load weights
-    if cfg.architecture in ['DeepSORT']:
-        if cfg.det_weights != 'None':
+    if cfg.architecture in ['DeepSORT', 'ByteTrack']:
         trainer.load_weights_sde(cfg.det_weights, cfg.reid_weights)
-        else:
-            trainer.load_weights_sde(None, cfg.reid_weights)
     else:
         trainer.load_weights(cfg.weights)
 

tools/infer_mot.py

@@ -34,9 +34,6 @@ from ppdet.engine import Tracker
 from ppdet.utils.check import check_gpu, check_version, check_config
 from ppdet.utils.cli import ArgsParser
 
-from ppdet.utils.logger import setup_logger
-logger = setup_logger('train')
-
 
 def parse_args():
     parser = ArgsParser()
@@ -94,11 +91,8 @@ def run(FLAGS, cfg):
     tracker = Tracker(cfg, mode='test')
 
     # load weights
-    if cfg.architecture in ['DeepSORT']:
-        if cfg.det_weights != 'None':
+    if cfg.architecture in ['DeepSORT', 'ByteTrack']:
         tracker.load_weights_sde(cfg.det_weights, cfg.reid_weights)
-        else:
-            tracker.load_weights_sde(None, cfg.reid_weights)
     else:
         tracker.load_weights_jde(cfg.weights)