search

InterDigitalInc / CompressAI

A PyTorch library and evaluation platform for end-to-end compression research
https://interdigitalinc.github.io/CompressAI/
BSD 3-Clause Clear License
1.19k stars 232 forks source link

Usage of QReLU #117

Closed Lithium07 closed 2 years ago

Lithium07 commented 2 years ago

Thanks for releasing the code of the video model ssf. I found that, in the Hyperprior model, the mean and scale use separate hyper decoders, and the scale decoder uses the QReLU function. The QReLU is originally designed for integral compression network in Google's paper. Is there any reason for using separate decoders and the QReLU (only used in the scale decoder) for the video codec? Hope to get your sharing and insight. Thanks a lot.

Lithium07 commented 2 years ago

And the bit_depth is set as 8, why not 1? image

chyomin06 commented 2 years ago

Hi,

I think it is good to have this on the Discussion page. Please, see this (https://github.com/InterDigitalInc/CompressAI/discussions/118)

Thank you.