guied-grad-camでのopencvのエラー文

前提・実現したいこと

guied-grad-camを使用したいのですが以下のエラー文が出ました。
調べたのですがユニコードが含まれていると出るなどがあったのですが
解決策がいまいちわかりません。
初心者なのでいまいちわかっていないので教えてください。

発生している問題・エラーメッセージ

 OpenCV(4.1.0) C:\projects\opencv-python\opencv\modules\imgcodecs\src\loadsave.cpp:667: error: (-215:Assertion failed) image.channels() == 1 || image.channels() == 3 || image.channels() == 4 in function 'cv::imwrite_'

該当のソースコード

python3
1 def target_category_loss(x, category_index, nb_classes):
2    return tf.multiply(x, K.one_hot([category_index], nb_classes))
3
4def target_category_loss_output_shape(input_shape):
5    return input_shape
6
7def normalize(x):
8    return x / (K.sqrt(K.mean(K.square(x))) + 1e-5)
9
10def load_image(path):
11        img_path = path
12    img = image.load_img(img_path, target_size=(224, 224))
13    x = image.img_to_array(img)
14    x = np.expand_dims(x, axis=0)
15    x = preprocess_input(x)
16    return x
17
18
19def register_gradient():
20    if "GuidedBackProp" not in ops._gradient_registry._registry:
21        @ops.RegisterGradient("GuidedBackProp")
22        def _GuidedBackProp(op, grad):
23            dtype = op.inputs[0].dtype
24            return grad * tf.cast(grad > 0., dtype) * \
25                tf.cast(op.inputs[0] > 0., dtype)
26
27
28def compile_saliency_function(model, activation_layer='block5_conv3'):
29    input_img = model.input
30    layer_dict = dict([(layer.name, layer) for layer in model.layers[1:]])
31    layer_output = layer_dict[activation_layer].output
32    max_output = K.max(layer_output, axis=3)
33    saliency = K.gradients(K.sum(max_output), input_img)[0]
34    return K.function([input_img, K.learning_phase()], [saliency])
35
36
37def modify_backprop(model, name):
38    g = tf.get_default_graph()
39    with g.gradient_override_map({'Relu': name}):
40        new_model = VGG16(weights='imagenet')
41    return new_model
42
43
44def deprocess_image(x):
45    if np.ndim(x) > 3:
46        x = np.squeeze(x)
47    x -= x.mean()
48    x /= (x.std() + 1e-5)
49    x *= 0.1
50
51    x += 0.5
52    x = np.clip(x, 0, 1)
53
54    x *= 255
55    if K.image_dim_ordering() == 'tf':
56        x = x.transpose((1, 2, 0))
57    x = np.clip(x, 0, 255).astype('uint8')
58    return x
59
60
61def grad_cam(input_model, image, category_index, layer_name):
62    nb_classes = 1000
63    target_layer = lambda x: target_category_loss(x, category_index, nb_classes)
64
65    x = input_model.layers[-1].output
66    x = Lambda(target_layer, output_shape=target_category_loss_output_shape)(x)
67    model = keras.models.Model(input_model.layers[0].input, x)
68
69    loss = K.sum(model.layers[-1].output)
70    conv_output = [l for l in model.layers if l.name is layer_name][0].output
71
72    grads = normalize(K.gradients(loss, conv_output)[0])
73    gradient_function = K.function([model.layers[0].input], [conv_output, grads])
74
75    output, grads_val = gradient_function([image])
76    output, grads_val = output[0, :], grads_val[0, :, :, :]
77
78    weights = np.mean(grads_val, axis = (0, 1))
79    cam = np.ones(output.shape[0 : 2], dtype = np.float32)
80
81    for i, w in enumerate(weights):
82        cam += w * output[:, :, i]
83
84    cam = cv2.resize(cam, (224, 224))
85    cam = np.maximum(cam, 0)
86    heatmap = cam / np.max(cam)
87
88    image = image[0, :]
89    image -= np.min(image)
90    image = np.minimum(image, 255)
91
92    cam = cv2.applyColorMap(np.uint8(255*heatmap), cv2.COLORMAP_JET)
93    cam = np.float32(cam) + np.float32(image)
94    cam = 255 * cam / np.max(cam)
95    return np.uint8(cam), heatmap
96
97preprocessed_input = load_image("./dog_cat.jpg")
98
99model = VGG16(weights='imagenet')
100
101predictions = model.predict(preprocessed_input)
102top_1 = decode_predictions(predictions)[0][0]
103print('Predicted class:')
104print('%s (%s) with probability %.2f' % (top_1[1], top_1[0], top_1[2]))
105
106predicted_class = np.argmax(predictions)
107print(predicted_class)
108
109cam, heatmap = grad_cam(model, preprocessed_input, predicted_class, "block5_conv3")
110
111cv2.imwrite("gradcam.jpg", cam)
112
113register_gradient()
114guided_model = modify_backprop(model, 'GuidedBackProp')
115saliency_fn = compile_saliency_function(guided_model)
116saliency = saliency_fn([preprocessed_input, 0])
117gradcam = saliency[0] * heatmap[..., np.newaxis]
118cv2.imwrite("guided_gradcam.jpg", deprocess_image(gradcam))
119