4 examples of 'keras model predict' in Python

15
def _predict(self, X, **kwargs):
16
    '''
17
    Keras returns class tuples (proba equivalent) so cast to single prediction
18
    '''
19
    return np.argmax(self.external_model.predict(X), axis=1)

31
def predict(args):
32
    # load the trained convolutional neural network
33
    print("[INFO] loading network...")
34
    model = load_model(args["model"])
35
    stride = args['stride']
36
    for n in range(len(TEST_SET)):
37
        path = TEST_SET[n]
38
        #load the image
39
        image = cv2.imread('./test/' + path)
40
        # pre-process the image for classification
41
        #image = image.astype("float") / 255.0
42
        #image = img_to_array(image)
43
        h,w,_ = image.shape
44
        padding_h = (h//stride + 1) * stride 
45
        padding_w = (w//stride + 1) * stride
46
        padding_img = np.zeros((padding_h,padding_w,3),dtype=np.uint8)
47
        padding_img[0:h,0:w,:] = image[:,:,:]
48
        padding_img = padding_img.astype("float") / 255.0
49
        padding_img = img_to_array(padding_img)
50
        print 'src:',padding_img.shape
51
        mask_whole = np.zeros((padding_h,padding_w),dtype=np.uint8)
52
        for i in range(padding_h//stride):
53
            for j in range(padding_w//stride):
54
                crop = padding_img[:3,i*stride:i*stride+image_size,j*stride:j*stride+image_size]
55
                _,ch,cw = crop.shape
56
                if ch != 256 or cw != 256:
57
                    print 'invalid size!'
58
                    continue
59
                    
60
                crop = np.expand_dims(crop, axis=0)
61
                #print 'crop:',crop.shape
62
                pred = model.predict_classes(crop,verbose=2)  
63
                pred = labelencoder.inverse_transform(pred[0])  
64
                #print (np.unique(pred))  
65
                pred = pred.reshape((256,256)).astype(np.uint8)
66
                #print 'pred:',pred.shape
67
                mask_whole[i*stride:i*stride+image_size,j*stride:j*stride+image_size] = pred[:,:]
68
69
        
70
        cv2.imwrite('./predict/pre'+str(n+1)+'.png',mask_whole[0:h,0:w])

47
@abstractmethod
48
def predict(self, x_test):
49
    """Return the results for the testing data predicted by the best neural architecture.
50
    
51
    Dependent on the results of the fit() function.
52
53
    Args:
54
        x_test: An instance of numpy.ndarray containing the testing data.
55
56
    Returns:
57
        A numpy.ndarray containing the predicted classes for x_test.
58
    """
59
    pass

372
def predict(self, x, batch_size=None, verbose=0, steps=None):
373
    target_bounding_boxes = numpy.zeros((x.shape[0], 1, 4))
374
375
    target_categories = numpy.zeros((x.shape[0], 1, self.n_categories))
376
377
    target_mask = numpy.zeros((1, 1, *self.mask_shape))
378
379
    target_metadata = numpy.array([[x.shape[1], x.shape[2], 1.0]])
380
381
    x = [
382
        target_bounding_boxes,
383
        target_categories,
384
        x,
385
        target_mask,
386
        target_metadata
387
    ]
388
389
    return super(RCNN, self).predict(x, batch_size, verbose, steps)