Variable "features" in net.py

[Hyeokreal]

Hi, yuamning-hu Thank you for openning implementation code.

I have a question

In the net.py line 325 - 354,

      for j in range(giters):
        feed_dict, features = self.get_training_feed_dict_and_states(iter)
        if iter == 0:
          feed_dict[self.lr_g] = 0
        feed_dict[self.progress] = progress
        _, g_loss, v_loss, fake_output, new_states = self.sess.run(
            [(self.opt_g, self.opt_v), self.g_loss, self.v_loss,
             self.fake_output, self.new_states],
            feed_dict=feed_dict,
            options=run_options,
            run_metadata=run_metadata)
        if self.cfg.supervised:
          ground_truth = feed_dict[self.ground_truth]
        else:
          ground_truth = None
        self.memory.replace_memory(
            self.memory.images_and_states_to_records(
                fake_output, new_states, features, ground_truth=ground_truth))
        v_loss_pool.append(v_loss)
        g_loss_pool.append(g_loss)

The variable named "features" seems always zeros. Is it meaningful variable ?

[yuanming-hu]

Thanks for the question. You are right features here is not useful.

During development, we tried to feed pretrained AlexNet features to the network, while it didn't help. (Then we got too lazy to remove it, but setting it to zero instead.)

[Hyeokreal]

Thank you so much.

I really love your research and i also think it is very useful to real-world tasks.

I have one more question. Have you ever tried GAIL structure for this photo-enhancement task ? I think GAIL structure could be simply implemented to your structure. So I wonder if you tried. If not, i'm gonna try it : )

[yuanming-hu]

Thanks! I haven't trailed GAIL but (based on the abstract of that paper,) it sounds very relevant to our work. Please let me posted if you find anything interesting!