IEICE TRANSACTIONS on Fundamentals
Adaptive Channel Scheduling for Acceleration and Fine Control of RNN-Based Image Compression
Summary :
The existing target-dependent scalable image compression network can control the target of the compressed images between the human visual system and the deep learning based classification task. However, in its RNN based structure controls the bit-rate through the number of iterations, where each iteration generates a fixed size of the bit stream. Therefore, a large number of iterations are required at the high BPP, and fine-grained image quality control is not supported at the low BPP. In this paper, we propose a novel RNN-based image compression model that can schedule the channel size per iteration, to reduce the number of iterations at the high BPP and fine-grained bit-rate control at the low BPP. To further enhance the efficiency, multiple network models for various channel sizes are combined into a single model using the slimmable network architecture. The experimental results show that the proposed method achieves comparable performance to the existing method with finer BPP adjustment, increases parameters by only 0.15% and reduces the average amount of computation by 40.4%.
- Publication
- IEICE TRANSACTIONS on Fundamentals Vol.E106-A No.9 pp.1211-1215
- Publication Date
- 2023/09/01
- Publicized
- 2023/06/13
- Online ISSN
- 1745-1337
- DOI
- 10.1587/transfun.2022IML0001
- Type of Manuscript
- Special Section LETTER (Special Section on Image Media Quality)
- Category
- Image
Authors
Sang Hoon KIM
Samsung Electronics,Sungkyunkwan University
Jong Hwan KO
Sungkyunkwan University
Keyword
Latest Issue
Copyrights notice of machine-translated contents
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Cite this
Copy
Sang Hoon KIM, Jong Hwan KO, "Adaptive Channel Scheduling for Acceleration and Fine Control of RNN-Based Image Compression" in IEICE TRANSACTIONS on Fundamentals,
vol. E106-A, no. 9, pp. 1211-1215, September 2023, doi: 10.1587/transfun.2022IML0001.
Abstract: The existing target-dependent scalable image compression network can control the target of the compressed images between the human visual system and the deep learning based classification task. However, in its RNN based structure controls the bit-rate through the number of iterations, where each iteration generates a fixed size of the bit stream. Therefore, a large number of iterations are required at the high BPP, and fine-grained image quality control is not supported at the low BPP. In this paper, we propose a novel RNN-based image compression model that can schedule the channel size per iteration, to reduce the number of iterations at the high BPP and fine-grained bit-rate control at the low BPP. To further enhance the efficiency, multiple network models for various channel sizes are combined into a single model using the slimmable network architecture. The experimental results show that the proposed method achieves comparable performance to the existing method with finer BPP adjustment, increases parameters by only 0.15% and reduces the average amount of computation by 40.4%.
URL: https://global.ieice.org/en_transactions/fundamentals/10.1587/transfun.2022IML0001/_p
Copy
@ARTICLE{e106-a_9_1211,
author={Sang Hoon KIM, Jong Hwan KO, },
journal={IEICE TRANSACTIONS on Fundamentals},
title={Adaptive Channel Scheduling for Acceleration and Fine Control of RNN-Based Image Compression},
year={2023},
volume={E106-A},
number={9},
pages={1211-1215},
abstract={The existing target-dependent scalable image compression network can control the target of the compressed images between the human visual system and the deep learning based classification task. However, in its RNN based structure controls the bit-rate through the number of iterations, where each iteration generates a fixed size of the bit stream. Therefore, a large number of iterations are required at the high BPP, and fine-grained image quality control is not supported at the low BPP. In this paper, we propose a novel RNN-based image compression model that can schedule the channel size per iteration, to reduce the number of iterations at the high BPP and fine-grained bit-rate control at the low BPP. To further enhance the efficiency, multiple network models for various channel sizes are combined into a single model using the slimmable network architecture. The experimental results show that the proposed method achieves comparable performance to the existing method with finer BPP adjustment, increases parameters by only 0.15% and reduces the average amount of computation by 40.4%.},
keywords={},
doi={10.1587/transfun.2022IML0001},
ISSN={1745-1337},
month={September},}
Copy
TY - JOUR
TI - Adaptive Channel Scheduling for Acceleration and Fine Control of RNN-Based Image Compression
T2 - IEICE TRANSACTIONS on Fundamentals
SP - 1211
EP - 1215
AU - Sang Hoon KIM
AU - Jong Hwan KO
PY - 2023
DO - 10.1587/transfun.2022IML0001
JO - IEICE TRANSACTIONS on Fundamentals
SN - 1745-1337
VL - E106-A
IS - 9
JA - IEICE TRANSACTIONS on Fundamentals
Y1 - September 2023
AB - The existing target-dependent scalable image compression network can control the target of the compressed images between the human visual system and the deep learning based classification task. However, in its RNN based structure controls the bit-rate through the number of iterations, where each iteration generates a fixed size of the bit stream. Therefore, a large number of iterations are required at the high BPP, and fine-grained image quality control is not supported at the low BPP. In this paper, we propose a novel RNN-based image compression model that can schedule the channel size per iteration, to reduce the number of iterations at the high BPP and fine-grained bit-rate control at the low BPP. To further enhance the efficiency, multiple network models for various channel sizes are combined into a single model using the slimmable network architecture. The experimental results show that the proposed method achieves comparable performance to the existing method with finer BPP adjustment, increases parameters by only 0.15% and reduces the average amount of computation by 40.4%.
ER -
English
