pushing code

Author: deshwalmahesh

.gitignore

@@ -0,0 +1,4 @@
+./traced_model.pt
+traced_model.pt
+./weights/yolov7x.pt
+/weights/yolov7x.pt

Demo.ipynb

@@ -0,0 +1,168 @@
+'''
+A Moduele which binds Yolov7 repo with Deepsort with modifications
+'''
+
+import os
+os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3' # comment out below line to enable tensorflow logging outputs
+import time
+import tensorflow as tf
+
+physical_devices = tf.config.experimental.list_physical_devices('GPU')
+if len(physical_devices) > 0:
+    tf.config.experimental.set_memory_growth(physical_devices[0], True)
+    
+import cv2
+import numpy as np
+import matplotlib.pyplot as plt
+
+from tensorflow.compat.v1 import ConfigProto # DeepSORT official implementation uses tf1.x so we have to do some modifications to avoid errors
+
+# deep sort imports
+from deep_sort import preprocessing, nn_matching
+from deep_sort.detection import Detection
+from deep_sort.tracker import Tracker
+
+# import from helpers
+from tracking_helpers import read_class_names, create_box_encoder
+from detection_helpers import *
+
+
+ # load configuration for object detector
+config = ConfigProto()
+config.gpu_options.allow_growth = True
+
+
+
+class YOLOv7_DeepSORT:
+    '''
+    Class to Wrap ANY detector  of YOLO type with DeepSORT
+    '''
+    def __init__(self, reID_model_path:str, detector, max_cosine_distance:float=0.4, nn_budget:float=None, nms_max_overlap:float=1.0,
+    coco_names_path:str ="./io_data/input/classes/coco.names",  ):
+        '''
+        args: 
+            reID_model_path: Path of the model which uses generates the embeddings for the cropped area for Re identification
+            detector: object of YOLO models or any model which gives you detections as [x1,y1,x2,y2,scores, class]
+            max_cosine_distance: Cosine Distance threshold for "SAME" person matching
+            nn_budget:  If not None, fix samples per class to at most this number. Removes the oldest samples when the budget is reached.
+            nms_max_overlap: Maximum NMs allowed for the tracker
+            coco_file_path: File wich contains the path to coco naames
+        '''
+        self.detector = detector
+        self.coco_names_path = coco_names_path
+        self.nms_max_overlap = nms_max_overlap
+        self.class_names = read_class_names()
+
+        # initialize deep sort
+        self.encoder = create_box_encoder(reID_model_path, batch_size=1)
+        metric = nn_matching.NearestNeighborDistanceMetric("cosine", max_cosine_distance, nn_budget) # calculate cosine distance metric
+        self.tracker = Tracker(metric) # initialize tracker
+
+
+    def track_video(self,video:str, output:str, skip_frames:int=0, show_live:bool=False, count_objects:bool=False, verbose:int = 0):
+        '''
+        Track any given webcam or video
+        args: 
+            video: path to input video or set to 0 for webcam
+            output: path to output video
+            skip_frames: Skip every nth frame. After saving the video, it'll have very visuals experience due to skipped frames
+            show_live: Whether to show live video tracking. Press the key 'q' to quit
+            count_objects: count objects being tracked on screen
+            verbose: print details on the screen allowed values 0,1,2
+        '''
+        try: # begin video capture
+            vid = cv2.VideoCapture(int(video))
+        except:
+            vid = cv2.VideoCapture(video)
+
+        out = None
+        if output: # get video ready to save locally if flag is set
+            width = int(vid.get(cv2.CAP_PROP_FRAME_WIDTH))  # by default VideoCapture returns float instead of int
+            height = int(vid.get(cv2.CAP_PROP_FRAME_HEIGHT))
+            fps = int(vid.get(cv2.CAP_PROP_FPS))
+            codec = cv2.VideoWriter_fourcc(*"XVID")
+            out = cv2.VideoWriter(output, codec, fps, (width, height))
+
+        frame_num = 0
+        while True: # while video is running
+            return_value, frame = vid.read()
+            if not return_value:
+                print('Video has ended or failed!')
+                break
+            frame_num +=1
+
+            if skip_frames and not frame_num % skip_frames: continue # skip every nth frame. When every frame is not important, you can use this to fasten the process
+            if verbose >= 1:start_time = time.time()
+
+            # -----------------------------------------PUT ANY DETECTION MODEL HERE -----------------------------------------------------------------
+            yolo_dets = self.detector.detect(frame.copy(), plot_bb = False)  # Get the detections
+            frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
+            
+            bboxes = yolo_dets[:,:4]
+            bboxes[:,2] = bboxes[:,2] - bboxes[:,0] # convert from xyxy to xywh
+            bboxes[:,3] = bboxes[:,3] - bboxes[:,1]
+
+            scores = yolo_dets[:,4]
+            classes = yolo_dets[:,-1]
+            num_objects = bboxes.shape[0]
+            # ---------------------------------------- DETECTION PART COMPLETED ---------------------------------------------------------------------
+            
+            names = []
+            for i in range(num_objects): # loop through objects and use class index to get class name
+                class_indx = int(classes[i])
+                class_name = self.class_names[class_indx]
+                names.append(class_name)
+
+            names = np.array(names)
+            count = len(names)
+
+            if count_objects:
+                cv2.putText(frame, "Objects being tracked: {}".format(count), (5, 35), cv2.FONT_HERSHEY_COMPLEX_SMALL, 1.5, (0, 0, 0), 2)
+
+            # ---------------------------------- DeepSORT tacker work starts here ------------------------------------------------------------
+            features = self.encoder(frame, bboxes) # encode detections and feed to tracker. [No of BB / detections per frame, embed_size]
+            detections = [Detection(bbox, score, class_name, feature) for bbox, score, class_name, feature in zip(bboxes, scores, names, features)] # [No of BB per frame] deep_sort.detection.Detection object
+
+            cmap = plt.get_cmap('tab20b') #initialize color map
+            colors = [cmap(i)[:3] for i in np.linspace(0, 1, 20)]
+
+            boxs = np.array([d.tlwh for d in detections])  # run non-maxima supression below
+            scores = np.array([d.confidence for d in detections])
+            classes = np.array([d.class_name for d in detections])
+            indices = preprocessing.non_max_suppression(boxs, classes, self.nms_max_overlap, scores)
+            detections = [detections[i] for i in indices]       
+
+            self.tracker.predict()  # Call the tracker
+            self.tracker.update(detections) #  updtate using Kalman Gain
+
+            for track in self.tracker.tracks:  # update new findings AKA tracks
+                if not track.is_confirmed() or track.time_since_update > 1:
+                    continue 
+                bbox = track.to_tlbr()
+                class_name = track.get_class()
+        
+                color = colors[int(track.track_id) % len(colors)]  # draw bbox on screen
+                color = [i * 255 for i in color]
+                cv2.rectangle(frame, (int(bbox[0]), int(bbox[1])), (int(bbox[2]), int(bbox[3])), color, 2)
+                cv2.rectangle(frame, (int(bbox[0]), int(bbox[1]-30)), (int(bbox[0])+(len(class_name)+len(str(track.track_id)))*17, int(bbox[1])), color, -1)
+                cv2.putText(frame, class_name + " : " + str(track.track_id),(int(bbox[0]), int(bbox[1]-11)),0, 0.6, (255,255,255),1, lineType=cv2.LINE_AA)    
+
+                if verbose == 2:
+                    print("Tracker ID: {}, Class: {},  BBox Coords (xmin, ymin, xmax, ymax): {}".format(str(track.track_id), class_name, (int(bbox[0]), int(bbox[1]), int(bbox[2]), int(bbox[3]))))
+                    
+            # -------------------------------- Tracker work ENDS here -----------------------------------------------------------------------
+            if verbose >= 1:
+                fps = 1.0 / (time.time() - start_time) # calculate frames per second of running detections
+                if not count_objects: print(f"Processed frame no: {frame_num} || Current FPS: {round(fps,2)}")
+                else: print(f"Processed frame no: {frame_num} || Current FPS: {round(fps,2)} || Objects tracked: {count}")
+            
+            result = np.asarray(frame)
+            result = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR)
+            
+            if output: out.write(result) # save output video
+
+            if show_live:
+                cv2.imshow("Output Video", result)
+                if cv2.waitKey(1) & 0xFF == ord('q'): break
+        
+        cv2.destroyAllWindows()

IO_data/input/images/horses.jpg

@@ -0,0 +1,49 @@
+# parameters
+nc: 80  # number of classes
+depth_multiple: 1.0  # model depth multiple
+width_multiple: 1.0  # layer channel multiple
+
+# anchors
+anchors:
+  - [12,16, 19,36, 40,28]  # P3/8
+  - [36,75, 76,55, 72,146]  # P4/16
+  - [142,110, 192,243, 459,401]  # P5/32
+
+# CSP-ResNet backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Stem, [128]],  # 0-P1/2
+   [-1, 3, ResCSPC, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 2-P3/8
+   [-1, 4, ResCSPC, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 4-P3/8
+   [-1, 6, ResCSPC, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 6-P3/8
+   [-1, 3, ResCSPC, [1024]],  # 7
+  ]
+
+# CSP-Res-PAN head
+head:
+  [[-1, 1, SPPCSPC, [512]], # 8
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [5, 1, Conv, [256, 1, 1]], # route backbone P4
+   [[-1, -2], 1, Concat, [1]],
+   [-1, 2, ResCSPB, [256]], # 13
+   [-1, 1, Conv, [128, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [3, 1, Conv, [128, 1, 1]], # route backbone P3
+   [[-1, -2], 1, Concat, [1]],
+   [-1, 2, ResCSPB, [128]], # 18
+   [-1, 1, Conv, [256, 3, 1]],
+   [-2, 1, Conv, [256, 3, 2]],
+   [[-1, 13], 1, Concat, [1]],  # cat
+   [-1, 2, ResCSPB, [256]], # 22
+   [-1, 1, Conv, [512, 3, 1]],
+   [-2, 1, Conv, [512, 3, 2]],
+   [[-1, 8], 1, Concat, [1]],  # cat
+   [-1, 2, ResCSPB, [512]], # 26
+   [-1, 1, Conv, [1024, 3, 1]],
+
+   [[19,23,27], 1, IDetect, [nc, anchors]],   # Detect(P3, P4, P5)
+  ]

IO_data/input/video/street.mp4

@@ -0,0 +1,49 @@
+# parameters
+nc: 80  # number of classes
+depth_multiple: 1.0  # model depth multiple
+width_multiple: 1.0  # layer channel multiple
+
+# anchors
+anchors:
+  - [12,16, 19,36, 40,28]  # P3/8
+  - [36,75, 76,55, 72,146]  # P4/16
+  - [142,110, 192,243, 459,401]  # P5/32
+
+# CSP-ResNeXt backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Stem, [128]],  # 0-P1/2
+   [-1, 3, ResXCSPC, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 2-P3/8
+   [-1, 4, ResXCSPC, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 4-P3/8
+   [-1, 6, ResXCSPC, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 6-P3/8
+   [-1, 3, ResXCSPC, [1024]],  # 7
+  ]
+
+# CSP-ResX-PAN head
+head:
+  [[-1, 1, SPPCSPC, [512]], # 8
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [5, 1, Conv, [256, 1, 1]], # route backbone P4
+   [[-1, -2], 1, Concat, [1]],
+   [-1, 2, ResXCSPB, [256]], # 13
+   [-1, 1, Conv, [128, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [3, 1, Conv, [128, 1, 1]], # route backbone P3
+   [[-1, -2], 1, Concat, [1]],
+   [-1, 2, ResXCSPB, [128]], # 18
+   [-1, 1, Conv, [256, 3, 1]],
+   [-2, 1, Conv, [256, 3, 2]],
+   [[-1, 13], 1, Concat, [1]],  # cat
+   [-1, 2, ResXCSPB, [256]], # 22
+   [-1, 1, Conv, [512, 3, 1]],
+   [-2, 1, Conv, [512, 3, 2]],
+   [[-1, 8], 1, Concat, [1]],  # cat
+   [-1, 2, ResXCSPB, [512]], # 26
+   [-1, 1, Conv, [1024, 3, 1]],
+
+   [[19,23,27], 1, IDetect, [nc, anchors]],   # Detect(P3, P4, P5)
+  ]

IO_data/output/street.avi

@@ -0,0 +1,52 @@
+# parameters
+nc: 80  # number of classes
+depth_multiple: 1.33  # model depth multiple
+width_multiple: 1.25  # layer channel multiple
+
+# anchors
+anchors:
+  - [12,16, 19,36, 40,28]  # P3/8
+  - [36,75, 76,55, 72,146]  # P4/16
+  - [142,110, 192,243, 459,401]  # P5/32
+
+# CSP-Darknet backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [32, 3, 1]],  # 0
+   [-1, 1, Conv, [64, 3, 2]],  # 1-P1/2
+   [-1, 1, Bottleneck, [64]],
+   [-1, 1, Conv, [128, 3, 2]],  # 3-P2/4
+   [-1, 2, BottleneckCSPC, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 5-P3/8
+   [-1, 8, BottleneckCSPC, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 7-P4/16
+   [-1, 8, BottleneckCSPC, [512]],
+   [-1, 1, Conv, [1024, 3, 2]], # 9-P5/32
+   [-1, 4, BottleneckCSPC, [1024]],  # 10
+  ]
+
+# CSP-Dark-PAN head
+head:
+  [[-1, 1, SPPCSPC, [512]], # 11
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [8, 1, Conv, [256, 1, 1]], # route backbone P4
+   [[-1, -2], 1, Concat, [1]],
+   [-1, 2, BottleneckCSPB, [256]], # 16 
+   [-1, 1, Conv, [128, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [6, 1, Conv, [128, 1, 1]], # route backbone P3
+   [[-1, -2], 1, Concat, [1]],
+   [-1, 2, BottleneckCSPB, [128]], # 21
+   [-1, 1, Conv, [256, 3, 1]],
+   [-2, 1, Conv, [256, 3, 2]],
+   [[-1, 16], 1, Concat, [1]],  # cat
+   [-1, 2, BottleneckCSPB, [256]], # 25
+   [-1, 1, Conv, [512, 3, 1]],
+   [-2, 1, Conv, [512, 3, 2]],
+   [[-1, 11], 1, Concat, [1]],  # cat
+   [-1, 2, BottleneckCSPB, [512]], # 29
+   [-1, 1, Conv, [1024, 3, 1]],
+
+   [[22,26,30], 1, IDetect, [nc, anchors]],   # Detect(P3, P4, P5)
+  ]

bridge_wrapper.py

@@ -0,0 +1,52 @@
+# parameters
+nc: 80  # number of classes
+depth_multiple: 1.0  # model depth multiple
+width_multiple: 1.0  # layer channel multiple
+
+# anchors
+anchors:
+  - [12,16, 19,36, 40,28]  # P3/8
+  - [36,75, 76,55, 72,146]  # P4/16
+  - [142,110, 192,243, 459,401]  # P5/32
+
+# CSP-Darknet backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [32, 3, 1]],  # 0
+   [-1, 1, Conv, [64, 3, 2]],  # 1-P1/2
+   [-1, 1, Bottleneck, [64]],
+   [-1, 1, Conv, [128, 3, 2]],  # 3-P2/4
+   [-1, 2, BottleneckCSPC, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 5-P3/8
+   [-1, 8, BottleneckCSPC, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 7-P4/16
+   [-1, 8, BottleneckCSPC, [512]],
+   [-1, 1, Conv, [1024, 3, 2]], # 9-P5/32
+   [-1, 4, BottleneckCSPC, [1024]],  # 10
+  ]
+
+# CSP-Dark-PAN head
+head:
+  [[-1, 1, SPPCSPC, [512]], # 11
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [8, 1, Conv, [256, 1, 1]], # route backbone P4
+   [[-1, -2], 1, Concat, [1]],
+   [-1, 2, BottleneckCSPB, [256]], # 16 
+   [-1, 1, Conv, [128, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [6, 1, Conv, [128, 1, 1]], # route backbone P3
+   [[-1, -2], 1, Concat, [1]],
+   [-1, 2, BottleneckCSPB, [128]], # 21
+   [-1, 1, Conv, [256, 3, 1]],
+   [-2, 1, Conv, [256, 3, 2]],
+   [[-1, 16], 1, Concat, [1]],  # cat
+   [-1, 2, BottleneckCSPB, [256]], # 25
+   [-1, 1, Conv, [512, 3, 1]],
+   [-2, 1, Conv, [512, 3, 2]],
+   [[-1, 11], 1, Concat, [1]],  # cat
+   [-1, 2, BottleneckCSPB, [512]], # 29
+   [-1, 1, Conv, [1024, 3, 1]],
+
+   [[22,26,30], 1, IDetect, [nc, anchors]],   # Detect(P3, P4, P5)
+  ]