sanity_checks_saliency
sanity_checks_saliency copied to clipboard
question about logitop
Dear Author
I read your paper, it is valuable work! Also thanks for sharing the code notebook. I was trying to follow your notebook on my own pretrained network (train with tf1)
I used y_logits = tf.layers.dense(xx,xx, name='logit') for computing the logit, while in your train_cnn_models.py you used y_logits = tf.matmul(h_fc1, W_fc2) + b_fc2. So in your cnn_mnist_cascading_randomization.ipynb you used y_logits_op = tf.get_collection('logits')[0] to get the logit op, in my case, i should use the "logit/BiasAdd" or "logit/MatMul" as the y_logits_op?
Thanks a lot!