COMPASS: High-Efficiency Deep Image Compression with Arbitrary-scale Spatial Scalability

Jongmin Park,1 Jooyoung Lee,2 Munchurl Kim*,1
1Korea Advanced Institute of Science and Technology, KAIST
2Electronics and Telecommunications Research Institute, ETRI
ICCV 2023
*Corresponding author.
Paper Code
Image Description

Visual comparison results of spatially scalable image compression methods for arbitrary scale factors: The image of Layer 2 is reconstructed from the image of Layer 1 with scale factor 2.0× for each method. The ‘acc. bits’ indicates the accumulated bits up to the corresponding layers.

Abstract

Recently, neural network (NN)-based image compression studies have actively been made and has shown impressive performance in comparison to traditional methods. However, most of the works have focused on non-scalable image compression (single-layer coding) while spatially scalable image compression has drawn less attention although it has many applications. In this paper, we propose a novel NN-based spatially scalable image compression method, called COMPASS, which supports arbitrary-scale spatial scalability. Our proposed COMPASS has a very flexible structure where the number of layers and their respective scale factors can be arbitrarily determined during inference. To reduce the spatial redundancy between adjacent layers for arbitrary scale factors, our COMPASS adopts an inter-layer arbitrary scale prediction method, called LIFF, based on implicit neural representation. We propose a combined RD loss function to effectively train multiple layers. Experimental results show that our COMPASS achieves BD-rate gain of -58.33% and -47.17% at maximum compared to SHVC and the state-of-the-art NN-based spatially scalable image compression method, respectively, for various combinations of scale factors. Our COMPASS also shows comparable or even better coding efficiency than the single-layer coding for various scale factors.

Visual results for multi-layer scalable coding

Visual results of our COMPASS for multi-layer scalable coding greater than three layers. We consider the total number of layers with three, four, five, six and nine, and compare the PSNR values and accumulated bits in each layer. As shown, our COMPASS shows faithful reconstructions over various layers.

Visual Comparison with a fixed scale factor of 2

Visual comparison results for images in Kodak Lossless True Color Image dataset with a fixed scale factor of 2. The ‘acc. bits’ indicates the accumulated bits up to the final EL. We match the accumulated bits among the compared methods as much as possible.

Video Presentation

ICCV Poster

BibTeX

@misc{park2023compass,
        title={COMPASS: High-Efficiency Deep Image Compression with Arbitrary-scale Spatial Scalability}, 
        author={Jongmin Park and Jooyoung Lee and Munchurl Kim},
        year={2023},
        eprint={2309.07926},
        archivePrefix={arXiv},
        primaryClass={eess.IV}
  }