RECORD
A Compression Objective and a Cycle Loss for Neural Image Compression
Image and Video Processing · Submitted 24 May 2019
abstract compiled · 2026-09-23 · no language-model review
The authors' main claim In particular, MSE-only training favors low image-domain distortion, whereas cycle loss training favors high perceptual quality.
The compiled record
The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.
Open the compiled record
The abstract, sentence by sentence
In this manuscript we propose two objective terms for neural image compression: a compression objective and a cycle loss. These terms are applied on the encoder output of an autoencoder and are used in combination with reconstruction losses. The compression objective encourages sparsity and low entropy in the activations. The cycle loss term represents the distortion between encoder outputs computed from the original image and from the reconstructed image (code-domain distortion). We train different autoencoders by using the compression objective in combination with different losses: a) MSE, b) MSE and MSSSIM, c) MSE, MS-SSIM and cycle loss. We observe that images encoded by these differently-trained autoencoders fall into different points of the perception-distortion curve (while having similar bit-rates). In particular, MSE-only training favors low image-domain distortion, whereas cycle loss training favors high perceptual quality.
background contribution method result interpretation main claim
| # | sentence | role | the authors' own work | states the main conclusion | strength |
|---|---|---|---|---|---|
| s0 | In this manuscript we propose two objective terms for neural image compression: a compression objective and a cycle loss. | contribution 1.00 | 0.89 | 0.08 | asserted 0.99 |
| s1 | These terms are applied on the encoder output of an autoencoder and are used in combination with reconstruction losses. | method 0.94 | 0.75 | 0.03 | asserted 1.00 |
| s2 | The compression objective encourages sparsity and low entropy in the activations. | method 0.77 | 0.36 | 0.04 | asserted 1.00 |
| s3 | The cycle loss term represents the distortion between encoder outputs computed from the original image and from the reconstructed image (code-domain distortion). | method 0.75 | 0.59 | 0.05 | asserted 1.00 |
| s4 | We train different autoencoders by using the compression objective in combination with different losses: a) MSE, b) MSE and MSSSIM, c) MSE, MS-SSIM and cycle loss. | method 0.98 | 0.94 | 0.05 | asserted 0.97 |
| s5 | We observe that images encoded by these differently-trained autoencoders fall into different points of the perception-distortion curve (while having similar bit-rates). | result 1.00 | 0.95 | 0.31 | asserted 0.99 |
| s6 | In particular, MSE-only training favors low image-domain distortion, whereas cycle loss training favors high perceptual quality. | interpretation 0.57 | 0.95 | 0.54 | asserted 0.99 |
What was examined, and what was not
- arXiv listing abstract v1
- full text
- figures and tables
- references and citing sentences
- relations to other papers
- P abstract
- S sections
- T body sentences
- C citations
- R cross-paper relations
The paper
Original abstract
In this manuscript we propose two objective terms for neural image compression: a compression objective and a cycle loss. These terms are applied on the encoder output of an autoencoder and are used in combination with reconstruction losses. The compression objective encourages sparsity and low entropy in the activations. The cycle loss term represents the distortion between encoder outputs computed from the original image and from the reconstructed image (code-domain distortion). We train different autoencoders by using the compression objective in combination with different losses: a) MSE, b) MSE and MSSSIM, c) MSE, MS-SSIM and cycle loss. We observe that images encoded by these differently-trained autoencoders fall into different points of the perception-distortion curve (while having similar bit-rates). In particular, MSE-only training favors low image-domain distortion, whereas cycle loss training favors high perceptual quality.
Human review
How the literature uses it
What this paper does with the arXiv papers it cites
What it cites · 1 sentence without a confident label
Cites as background 1 paper
-
In [ref] , [ref] , a generative adversarial network (GAN) was used in order to obtain images with better visual quality.
Generative Adversarial Networks for Extreme Learned Image Compression · Related Work · 0.80
Discussion
Know something about this paper? Leave a comment, or highlight any passage above to comment on it.