Encode/decode images with higher resolution than the autoencoder input

[Leonardo Augusto]

Hello everyone,

I wanted to know how did Johannes Ballé encoded and decoded images with  752 × 376 pixels with the autoencoder presented on the End-to-end Optimized Image Compression paper.
Is the autoencoder trained with an input of that size? Or is it possible to compress/decompress an image with an autoencoder trained with a smaller input?

[Fabian Mentzer]

The autoencoder is typically fully convolutional (made from 2D Conv layers or pointwise operations) and can thus be applied to any resolution, as long as you pad to a mutiple of the downscaling factor.

--
You received this message because you are subscribed to the Google Groups "TensorFlow Compression" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tensorflow-compre...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/tensorflow-compression/b8f0d0df-cc77-4d57-96bf-8c0b688e34fcn%40googlegroups.com.

[Leonardo Augusto]

Ohh, I see. I also found this stackoverflow post that helped me understand the inputs of the layers as well https://stackoverflow.com/questions/60523012/train-network-in-keras-consisting-only-of-conv2d-layers.
I didn't knew you could set the inputs as None.

Thanks for the quick response!