add frame-skip to boost inference (#6383)

deploy/pipeline/config/infer_cfg_pphuman.yml

@@ -50,6 +50,7 @@ ID_BASED_DETACTION:
   basemode: "idbased"
   threshold: 0.6
   display_frames: 80
+  skip_frame_num: 2
   enable: False
 
 ID_BASED_CLSACTION:
@@ -58,6 +59,7 @@ ID_BASED_CLSACTION:
   basemode: "idbased"
   threshold: 0.8
   display_frames: 80
+  skip_frame_num: 2
   enable: False
 
 REID:

deploy/pipeline/pipeline.py

@@ -342,7 +342,9 @@ class PipePredictor(object):
                 basemode = idbased_detaction_cfg['basemode']
                 threshold = idbased_detaction_cfg['threshold']
                 display_frames = idbased_detaction_cfg['display_frames']
+                skip_frame_num = idbased_detaction_cfg['skip_frame_num']
                 self.modebase[basemode] = True
+
                 self.det_action_predictor = DetActionRecognizer(
                     model_dir,
                     device,
@@ -355,7 +357,8 @@ class PipePredictor(object):
                     cpu_threads,
                     enable_mkldnn,
                     threshold=threshold,
-                    display_frames=display_frames)
+                    display_frames=display_frames,
+                    skip_frame_num=skip_frame_num)
                 self.det_action_visual_helper = ActionVisualHelper(1)
 
             if self.with_idbased_clsaction:
@@ -366,6 +369,8 @@ class PipePredictor(object):
                 threshold = idbased_clsaction_cfg['threshold']
                 self.modebase[basemode] = True
                 display_frames = idbased_clsaction_cfg['display_frames']
+                skip_frame_num = idbased_clsaction_cfg['skip_frame_num']
+
                 self.cls_action_predictor = ClsActionRecognizer(
                     model_dir,
                     device,
@@ -378,7 +383,8 @@ class PipePredictor(object):
                     cpu_threads,
                     enable_mkldnn,
                     threshold=threshold,
-                    display_frames=display_frames)
+                    display_frames=display_frames,
+                    skip_frame_num=skip_frame_num)
                 self.cls_action_visual_helper = ActionVisualHelper(1)
 
             if self.with_skeleton_action:

deploy/pipeline/pphuman/action_infer.py

@@ -279,7 +279,11 @@ class DetActionRecognizer(object):
         cpu_threads (int): cpu threads
         enable_mkldnn (bool): whether to open MKLDNN
         threshold (float): The threshold of score for action feature object detection.
-        display_frames (int): The duration for corresponding detected action. 
+        display_frames (int): The duration for corresponding detected action.
+        skip_frame_num (int): The number of frames for interval prediction. A skipped frame will 
+            reuse the result of its last frame. If it is set to 0, no frame will be skipped. Default
+            is 0.
+
     """
 
     def __init__(self,
@@ -295,7 +299,8 @@ class DetActionRecognizer(object):
                  enable_mkldnn=False,
                  output_dir='output',
                  threshold=0.5,
-                 display_frames=20):
+                 display_frames=20,
+                 skip_frame_num=0):
         super(DetActionRecognizer, self).__init__()
         self.detector = Detector(
             model_dir=model_dir,
@@ -313,10 +318,21 @@ class DetActionRecognizer(object):
         self.threshold = threshold
         self.frame_life = display_frames
         self.result_history = {}
+        self.skip_frame_num = skip_frame_num
+        self.skip_frame_cnt = 0
+        self.id_in_last_frame = []
 
     def predict(self, images, mot_result):
-        det_result = self.detector.predict_image(images, visual=False)
-        result = self.postprocess(det_result, mot_result)
+        if self.skip_frame_cnt == 0 or (not self.check_id_is_same(mot_result)):
+            det_result = self.detector.predict_image(images, visual=False)
+            result = self.postprocess(det_result, mot_result)
+        else:
+            result = self.reuse_result(mot_result)
+
+        self.skip_frame_cnt += 1
+        if self.skip_frame_cnt >= self.skip_frame_num:
+            self.skip_frame_cnt = 0
+
         return result
 
     def postprocess(self, det_result, mot_result):
@@ -343,10 +359,11 @@ class DetActionRecognizer(object):
             if valid_boxes.shape[0] >= 1:
                 action_ret['class'] = valid_boxes[0, 0]
                 action_ret['score'] = valid_boxes[0, 1]
-                self.result_history[tracker_id] = [0, self.frame_life]
+                self.result_history[
+                    tracker_id] = [0, self.frame_life, valid_boxes[0, 1]]
             else:
-                history_det, life_remain = self.result_history.get(tracker_id,
-                                                                   [1, 0])
+                history_det, life_remain, history_score = self.result_history.get(
+                    tracker_id, [1, self.frame_life, -1.0])
                 action_ret['class'] = history_det
                 action_ret['score'] = -1.0
                 life_remain -= 1
@@ -354,10 +371,48 @@ class DetActionRecognizer(object):
                     del (self.result_history[tracker_id])
                 elif tracker_id in self.result_history:
                     self.result_history[tracker_id][1] = life_remain
+                else:
+                    self.result_history[tracker_id] = [
+                        history_det, life_remain, history_score
+                    ]
+
+            mot_id.append(tracker_id)
+            act_res.append(action_ret)
+        result = list(zip(mot_id, act_res))
+        self.id_in_last_frame = mot_id
+
+        return result
+
+    def check_id_is_same(self, mot_result):
+        mot_bboxes = mot_result.get('boxes')
+        for idx in range(len(mot_bboxes)):
+            tracker_id = mot_bboxes[idx, 0]
+            if tracker_id not in self.id_in_last_frame:
+                return False
+        return True
+
+    def reuse_result(self, mot_result):
+        # This function reusing previous results of the same ID directly.
+        mot_bboxes = mot_result.get('boxes')
+
+        mot_id = []
+        act_res = []
+
+        for idx in range(len(mot_bboxes)):
+            tracker_id = mot_bboxes[idx, 0]
+            history_cls, life_remain, history_score = self.result_history.get(
+                tracker_id, [1, 0, -1.0])
+
+            life_remain -= 1
+            if tracker_id in self.result_history:
+                self.result_history[tracker_id][1] = life_remain
 
+            action_ret = {'class': history_cls, 'score': history_score}
             mot_id.append(tracker_id)
             act_res.append(action_ret)
+
         result = list(zip(mot_id, act_res))
+        self.id_in_last_frame = mot_id
 
         return result
 
@@ -378,6 +433,9 @@ class ClsActionRecognizer(AttrDetector):
         enable_mkldnn (bool): whether to open MKLDNN
         threshold (float): The threshold of score for action feature object detection.
         display_frames (int): The duration for corresponding detected action. 
+        skip_frame_num (int): The number of frames for interval prediction. A skipped frame will 
+            reuse the result of its last frame. If it is set to 0, no frame will be skipped. Default
+            is 0.
     """
 
     def __init__(self,
@@ -393,7 +451,8 @@ class ClsActionRecognizer(AttrDetector):
                  enable_mkldnn=False,
                  output_dir='output',
                  threshold=0.5,
-                 display_frames=80):
+                 display_frames=80,
+                 skip_frame_num=0):
         super(ClsActionRecognizer, self).__init__(
             model_dir=model_dir,
             device=device,
@@ -410,11 +469,22 @@ class ClsActionRecognizer(AttrDetector):
         self.threshold = threshold
         self.frame_life = display_frames
         self.result_history = {}
+        self.skip_frame_num = skip_frame_num
+        self.skip_frame_cnt = 0
+        self.id_in_last_frame = []
 
     def predict_with_mot(self, images, mot_result):
-        images = self.crop_half_body(images)
-        cls_result = self.predict_image(images, visual=False)["output"]
-        result = self.match_action_with_id(cls_result, mot_result)
+        if self.skip_frame_cnt == 0 or (not self.check_id_is_same(mot_result)):
+            images = self.crop_half_body(images)
+            cls_result = self.predict_image(images, visual=False)["output"]
+            result = self.match_action_with_id(cls_result, mot_result)
+        else:
+            result = self.reuse_result(mot_result)
+
+        self.skip_frame_cnt += 1
+        if self.skip_frame_cnt >= self.skip_frame_num:
+            self.skip_frame_cnt = 0
+
         return result
 
     def crop_half_body(self, images):
@@ -456,8 +526,8 @@ class ClsActionRecognizer(AttrDetector):
             # Current now,  class 0 is positive, class 1 is negative.
             if cls_id_res == 1 or (cls_id_res == 0 and
                                    cls_score_res < self.threshold):
-                history_cls, life_remain = self.result_history.get(tracker_id,
-                                                                   [1, 0])
+                history_cls, life_remain, history_score = self.result_history.get(
+                    tracker_id, [1, self.frame_life, -1.0])
                 cls_id_res = history_cls
                 cls_score_res = 1 - cls_score_res
                 life_remain -= 1
@@ -465,13 +535,51 @@ class ClsActionRecognizer(AttrDetector):
                     del (self.result_history[tracker_id])
                 elif tracker_id in self.result_history:
                     self.result_history[tracker_id][1] = life_remain
+                else:
+                    self.result_history[
+                        tracker_id] = [cls_id_res, life_remain, cls_score_res]
             else:
-                self.result_history[tracker_id] = [cls_id_res, self.frame_life]
+                self.result_history[
+                    tracker_id] = [cls_id_res, self.frame_life, cls_score_res]
 
             action_ret = {'class': cls_id_res, 'score': cls_score_res}
             mot_id.append(tracker_id)
             act_res.append(action_ret)
         result = list(zip(mot_id, act_res))
+        self.id_in_last_frame = mot_id
+
+        return result
+
+    def check_id_is_same(self, mot_result):
+        mot_bboxes = mot_result.get('boxes')
+        for idx in range(len(mot_bboxes)):
+            tracker_id = mot_bboxes[idx, 0]
+            if tracker_id not in self.id_in_last_frame:
+                return False
+        return True
+
+    def reuse_result(self, mot_result):
+        # This function reusing previous results of the same ID directly.
+        mot_bboxes = mot_result.get('boxes')
+
+        mot_id = []
+        act_res = []
+
+        for idx in range(len(mot_bboxes)):
+            tracker_id = mot_bboxes[idx, 0]
+            history_cls, life_remain, history_score = self.result_history.get(
+                tracker_id, [1, 0, -1.0])
+
+            life_remain -= 1
+            if tracker_id in self.result_history:
+                self.result_history[tracker_id][1] = life_remain
+
+            action_ret = {'class': history_cls, 'score': history_score}
+            mot_id.append(tracker_id)
+            act_res.append(action_ret)
+
+        result = list(zip(mot_id, act_res))
+        self.id_in_last_frame = mot_id
 
         return result