Quantized Decoder in Learned Image Compression for Deterministic Reconstruction

Koyuncu E, Solovyev T, Sauer J, Alshina E, Kaup A (2024)

Publication Language: English

Publication Type: Conference contribution, Conference Contribution

Publication year: 2024

Journal

2013 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

Publisher: Institute of Electrical and Electronics Engineers Inc.

Pages Range: 3985-3989

Conference Proceedings Title: IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

Event location: Seoul KR

DOI: 10.1109/ICASSP48485.2024.10448359

Open Access Link: https://arxiv.org/abs/2312.11209

Abstract

Learned image compression has a problem of non-bit-exact reconstruction due to different calculations of floating point arithmetic on different devices. This paper shows a method to achieve a deterministic reconstructed image by quantizing only the decoder of the learned image compression model. From the implementation perspective of an image codec, it is beneficial to have the results reproducible when decoded on different devices. In this paper, we study quantization of weights and activations without overflow of accumulator in all decoder subnetworks. We show that the results are bit-exact at the output, and the resulting BD-rate loss of quantization of decoder is 0.5% in the case of 16-bit weights and 16-bit activations, and 7.9% in the case of 8-bit weights and 16-bit activations.

Authors with CRIS profile

Esin Koyuncu Lehrstuhl für Multimediakommunikation und Signalverarbeitung (LMS) Andre Kaup Lehrstuhl für Multimediakommunikation und Signalverarbeitung (LMS)

Involved external institutions

HUAWEI TECHNOLOGIES Deutschland GmbH DE Germany (DE)

How to cite

APA:

Koyuncu, E., Solovyev, T., Sauer, J., Alshina, E., & Kaup, A. (2024). Quantized Decoder in Learned Image Compression for Deterministic Reconstruction. In IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) (pp. 3985-3989). Seoul, KR: Institute of Electrical and Electronics Engineers Inc..

MLA:

Koyuncu, Esin, et al. "Quantized Decoder in Learned Image Compression for Deterministic Reconstruction." Proceedings of the 2024 - IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), Seoul Institute of Electrical and Electronics Engineers Inc., 2024. 3985-3989.

BibTeX: Download