eduzhai > Applied Sciences > Engineering >

CompressNet Generative Compression at Extremely Low Bitrates

  • Save

... pages left unread,continue reading

Document pages: 9 pages

Abstract: Compressing images at extremely low bitrates (< 0.1 bpp) has always been achallenging task since the quality of reconstruction significantly reduces dueto the strong imposed constraint on the number of bits allocated for thecompressed data. With the increasing need to transfer large amounts of imageswith limited bandwidth, compressing images to very low sizes is a crucial task.However, the existing methods are not effective at extremely low bitrates. Toaddress this need, we propose a novel network called CompressNet which augmentsa Stacked Autoencoder with a Switch Prediction Network (SAE-SPN). This helps inthe reconstruction of visually pleasing images at these low bitrates (< 0.1bpp). We benchmark the performance of our proposed method on the Cityscapesdataset, evaluating over different metrics at extremely low bitrates to showthat our method outperforms the other state-of-the-art. In particular, at abitrate of 0.07, CompressNet achieves 22 lower Perceptual Loss and 55 lowerFrechet Inception Distance (FID) compared to the deep learning SOTA methods.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×