Tensorflow freeze model but backpropagate error and update input layer

2018-06-11 03:53:52

I try to explain my goal. With a trained model I want to select the output class and update the feeded image.

Assign an image to the 'input_layer'.

Forward and compute error of 'output_layer' against desired output/class.

Backpropagate the error to the 'input_layer' without updating the weights and biases of the net.

Update the input layer, the original image, with the backpropagated error.

Some hint?

You can use tf.gradients to back propagate to the input layer:

...
logits = run_net(image)
g = tf.gradients(logits[target_class], image)
image += g[0] * step
...

Good examples of doing this can be found in the Deep Dream demo code (see for example "Naive feature visualization" or "Multiscale image generation."

链接地址: http://www.djcxy.com/p/32062.html