Efficient Learning Based Sub-pixel Image Compression

Chunlei Cai, Guo Lu, Qiang Hu, Li Chen, Zhiyong Gao; The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2019, pp. 0-0

Abstract


In this paper, we propose an efficient learning based sub-pixel image compression algorithm. Our framework builds upon the previous variational auto-encoder architecture and reduces the computational complexity significantly. Specifically, we propose an end-to-end optimized image compression framework to utilize the powerful non-linear representation ability of neural networks. This framework follows the widely used variational auto-encoder architecture and is optimized based on the rate-distortion balance. More importantly, a sub-pixel image compression framework is exploited to reduce the spatial resolution of image and improve the inference speed. Experimental results demonstrate the effectiveness of our method. Compared with the baseline algorithm, our encoder is 2 times faster with negligible performance decrease. The decoding speed of our method for the CLIC dataset is 1.85 fps on GTX 1080Ti, which makes our codec one of the fastest learning based image compression algorithm.

Related Material


[pdf]
[bibtex]
@InProceedings{Cai_2019_CVPR_Workshops,
author = {Cai, Chunlei and Lu, Guo and Hu, Qiang and Chen, Li and Gao, Zhiyong},
title = {Efficient Learning Based Sub-pixel Image Compression},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2019}
}