JTan

Hi, Very sorry, misread your question. I didn’t consider those steps in my code. Unfortunately I don’t have much experience with information theory, but that does look interesting to me...

I think it shouldn’t actually affect the network ability and will just improve the compression, but not 100% sure. On Wed, 31 Jul 2019 at 4:33 pm, shellhuang1227 wrote: >...

If you read the original paper (https://arxiv.org/pdf/1804.02958.pdf), the upper bound on the bitrate is given by Eq. 5. Here dim (`w_hat`) is given by the number of channels C.

If you want to compress arbitrary images, train on a large dataset of *natural images* like ImageNet or the ADE20k dataset. The pretrained model was only trained on the Cityscapes...

Don't train using Cityscapes initially, just train using ADE20k. Make sure you pull the latest version, I fixed a couple of errors in the code. It should take a long...

I will look into the the first one, can you give more details about the second point?

IIRC, I manually overrode the gradient so that it is just the identity - this is the 'straight-through' estimator that works surprisingly well in practice. On Tue, May 25, 2021...

The batch size is limited by GPU memory. In recent papers about learned image compression the batch size is usually set to something low like 8 or 16, so I...

Hey, this looks good, unfortunately it's been a while since I looked at the codebase. I agree from the documentation that this should increase performance. Happy to accept a PR...

The dataset is pretty large, you may want to subsample the data as an additional preprocessing step and modify the config file to the new input size. Try 256 x...