First, I want to thank your wonderful code which is useful for my research. I have a question about the target_rate variable which controls the bpp: I did not find target_rate is used in any other code documents(maybe i just miss it). I wonder which part is changed between pre-train models with different regime and how target_rate is used? And if I want to continue increasing the bpp, did I only need to change number of latent_channels(or maybe the quantization precision in hyperprior.py). Any suggestion about the change need to do to increase bpp will be useful.
First, I want to thank your wonderful code which is useful for my research. I have a question about the target_rate variable which controls the bpp: I did not find target_rate is used in any other code documents(maybe i just miss it). I wonder which part is changed between pre-train models with different regime and how target_rate is used? And if I want to continue increasing the bpp, did I only need to change number of latent_channels(or maybe the quantization precision in hyperprior.py). Any suggestion about the change need to do to increase bpp will be useful.
Thanks a lot for your time.