How is referencenet(latents_ref_img, ref_timesteps, encoder_hidden_states) in Line 492 in train.py connected to the loss? 

There is no variable thats accepting the response of referencenet neither am I sending the sample to any other network. How Does it work?