I'm working on GradCam from my previous version of the code that works. However, after i updated tensorflow to version 2, the code is no longer work. My previous code is the following
def find_target_layer(model):
"""
This function finds the final convolutional layer in the network by
looping over the layers of the network in reverse order
Raises
------
ValueError
Report error if 4D layer is not found.
Returns
-------
TYPE
DESCRIPTION.
"""
for layer in reversed(model.layers):
if len(layer.output_shape) == 4:
return layer.name
raise ValueError("Could not find 4D layer. Cannot apply GradCAM")
def grad_cam(input_model, image, class_idx, layer_name, H=224, W=224):
"""
This function apply GradCAM method for visualizing input saliency.
Parameters
----------
input_model : Keras Model Object
Trained Model
image : 4D Numpy Array (1, H, W, 3)
Preprocessed input for a given model
class_idx : Integer
Class index
layer_name : String
Layer name of keras model to apply GradCAM
H : Integer, optional
Image height. The default is 224.
W : Integer, optional
Image width. The default is 224.
Returns
-------
cam : 2D Numpy Array (H, W)
Heatmap.
"""
target_layer = find_target_layer(input_model)
y_c = input_model.output[0, class_idx]
conv_output = input_model.get_layer(target_layer).output
grads = K.gradients(y_c, conv_output)[0]
gradient_function = K.function([input_model.input], [conv_output, grads])
output, grads_val = gradient_function([image])
output, grads_val = output[0, :], grads_val[0, :, :, :]
weights = np.mean(grads_val, axis=(0, 1))
cam = np.dot(output, weights)
# Process CAM
cam = cv2.resize(cam, (W, H), cv2.INTER_LINEAR)
cam = np.maximum(cam, 0)
cam = cam / cam.max()
return cam
I debugged line by line using densenet121 but it failed at the line grads = K.gradients(y_c, conv_output)[0]. The error message is
TypeError: Cannot convert a symbolic Keras input/output to a numpy array.
This error may indicate that you're trying to pass a symbolic value to a NumPy call,
which is not supported. Or, you may be trying to pass Keras symbolic inputs/outputs
to a TF API that does not register dispatching, preventing Keras from automatically
converting the API call to a lambda layer in the Functional Model.
May I have your suggestions how to fix this? I'm aware there are several tutorials work on GradCam, however, I still would like to learn more how to fix this code to gain more understanding.
question from:https://stackoverflow.com/questions/65921392/gradcam-for-keras-tensorflow-2
