How to deal with the file bpp and computing bpp?

LiuLei95 / PyTorch-Learned-Image-Compression-with-GMM-and-Attention

This repo is implementation for Learned Image Compression with Discretized Gaussian Mixture Likelihoods and Attention Modules in pytorch.

Apache License 2.0

65 stars 3 forks source link

How to deal with the file bpp and computing bpp? #15

Open xichuanhit opened 7 months ago

xichuanhit commented 7 months ago

The bpp computing by the output npz file of Decoder is not equalled with the bpp compute from the net ImageCompressor. The resource image is 1920*1920, file size is 2534.633 Kb. While I save the output of ImageCompressor.Decoder(), the npz file size is 162.515 Kb. The bpp computed through the file size is 0.3526801215277778, which is much higher than the bpp_feature computed by the forward of net ImageCompressor (bpp-feature:0.09008687734603882). How to deal with the difference?