output = Dense(num_classes, activation='softmax')(x)# Create the fine-tuned modelmodel = Model(inputs=base_model.input, outputs=output)# Compile the modelmodel.compile(optimizer=Adam(lr=0.001), loss='categorical_crossentropy', metrics=['accuracy'])# Fine-tune on sports action video ...
grads = opt.compute_gradients(total_loss) Apply gradients. apply_gradient_op = opt.apply_gradients(grads, global_step=global_step) #apply_gradient_op=opt.minimize(total_loss) #apply_gradient_op = tf.train.GradientDescentOptimizer(lr).minimize(total_loss,global_step=global_step) Add histograms ...
Moreover, why does the attention classifier not use an Arc-face like the global classifier, but use the standard softmax-cross-entropy? Is there any benefit of doing this way? 👍 1 Member andrefaraujo commented Sep 3, 2020 We found that pooling on the local descriptor directly can be...