Tensorflow Mask-RCNN（三）——实时 检测视频

参考：https://www.youtube.com/watch?v=lLM8oAsi32g

    import cv2
    import numpy as np
     
     
    def random_colors(N):
        np.random.seed(1)
        colors=[tuple(255*np.random.rand(3)) for _ in range(N)]
        return colors
     
    def apply_mask(image, mask, color, alpha=0.5):
        """Apply the given mask to the image.
        """
        for n, c in enumerate(color):
            image[:, :, n] = np.where(
                mask == 1,
                image[:, :, n] *(1 - alpha) + alpha * c,
                image[:, :, n]
            )
        return image
     
    def display_instances(image,boxes,masks,ids,names,scores):
        n_instances=boxes.shape[0]
        if not n_instances:
            print('No instances to display')
        else:
            assert boxes.shape[0] == masks.shape[-1] == ids.shape[0]
        
        colors=random_colors(n_instances)
        height, width = image.shape[:2]
        
        for i,color in enumerate(colors):
            if not np.any(boxes[i]):
                continue
            
            y1,x1,y2,x2=boxes[i]
            mask=masks[:,:,i]
            image=apply_mask(image,mask,color)
            image=cv2.rectangle(image,(x1,y1),(x2,y2),color,2)
            
            label=names[ids[i]]
            score=scores[i] if scores is not None else None
            
            caption='{}{:.2f}'.format(label,score) if score else label
            image=cv2.putText(
                image,caption,(x1,y1),cv2.FONT_HERSHEY_COMPLEX,0.7,color,2
            )
            
        return image
     
    if __name__=='__main__':
        import os
        import sys
        import random
        import math
        import skimage.io
        import time
        import utils
        #import model as modellib
        
        
        ROOT_DIR = os.path.abspath("../")
        sys.path.append(ROOT_DIR)
        from mrcnn import utils
        import mrcnn.model as modellib
     
     
        sys.path.append(os.path.join(ROOT_DIR, "samples/coco/"))  # To find local version
        import coco
        
     
        MODEL_DIR = os.path.join(ROOT_DIR, "logs")
        COCO_MODEL_PATH = os.path.join(ROOT_DIR, "mask_rcnn_coco.h5")
        if not os.path.exists(COCO_MODEL_PATH):
            print('cannot find coco_model')
            
        class InferenceConfig(coco.CocoConfig):
            GPU_COUNT = 1
            IMAGES_PER_GPU = 1
     
        config = InferenceConfig()
        config.display()
        
        model = modellib.MaskRCNN(
            mode="inference", model_dir=MODEL_DIR, config=config
        )
     
        # Load weights trained on MS-COCO
        model.load_weights(COCO_MODEL_PATH, by_name=True)
        class_names = ['BG', 'person', 'bicycle', 'car', 'motorcycle', 'airplane',
                   'bus', 'train', 'truck', 'boat', 'traffic light',
                   'fire hydrant', 'stop sign', 'parking meter', 'bench', 'bird',
                   'cat', 'dog', 'horse', 'sheep', 'cow', 'elephant', 'bear',
                   'zebra', 'giraffe', 'backpack', 'umbrella', 'handbag', 'tie',
                   'suitcase', 'frisbee', 'skis', 'snowboard', 'sports ball',
                   'kite', 'baseball bat', 'baseball glove', 'skateboard',
                   'surfboard', 'tennis racket', 'bottle', 'wine glass', 'cup',
                   'fork', 'knife', 'spoon', 'bowl', 'banana', 'apple',
                   'sandwich', 'orange', 'broccoli', 'carrot', 'hot dog', 'pizza',
                   'donut', 'cake', 'chair', 'couch', 'potted plant', 'bed',
                   'dining table', 'toilet', 'tv', 'laptop', 'mouse', 'remote',
                   'keyboard', 'cell phone', 'microwave', 'oven', 'toaster',
                   'sink', 'refrigerator', 'book', 'clock', 'vase', 'scissors',
                   'teddy bear', 'hair drier', 'toothbrush']
        
        
        capture=cv2.VideoCapture(0)
        capture.set(cv2.CAP_PROP_FRAME_WIDTH,1920)
        capture.set(cv2.CAP_PROP_FRAME_HEIGHT,1080)
        
        while True:
            ret,frame=capture.read()
            results=model.detect([frame],verbose=0)
            r=results[0]
            
            
            frame=display_instances(
                  frame,r['rois'], r['masks'], r['class_ids'],
                                class_names, r['scores']
            )
            
            cv2.imshow('frame',frame)
            if cv2.waitKey(1)&0xFF==ord('q'):
                break
           
        capture.release()
        cv2.destroyAllWindows()