Hi!
I have a question regarding how the quantized representation can be compressed in practice. In particular, I understand that entropy is used as an estimation of the "size" the compressed file should have. But how such file can be obtained at inference time? I have seen related work using, for example, lossless entropy coding algorithms or huffman coding, among others.

Thanks for your help!

Hi authors, thanks a lot for sharing the code.
I want to understand for HQA for MNIST, what does the hidden units for each layer denote?
In the code, I see the number of hidden units for each layer's encoder is [16, 16, 32, 64, 128]. I understand this to be the number of output channels of the middle layer of each 3-layer feedforward encoder network?
For decoder, its the number of output channels of the first layer in the decoder after quantization layer, which is [16, 64, 256, 512, 1024].
Is this the case?

Secondly, in Table 10 from the paper, the number of hidden layers units for each layer's encoder is [16, 16, 32, 48, 80], while for decoders, it's [16, 32, 48, 80, 128]. These don't match the numbers in the code, am I reading them incorrectly? Thanks!

Hey.
I am newbie in computer vision research. I want this code to try on face image reconstruction as you have mentioned in paper. I find it difficult to understand the paper in the first attempt. Any help or suggestions ?