分布式策略下的解码端增强图像压缩网络

doi:10.19665/j.issn1001-2400.20241012

摘要/Abstract

摘要：

随着多媒体的快速发展,大规模的图像数据对网络带宽和存储空间造成了巨大压力。目前,基于深度学习的图像压缩方法仍然存在重建图像有压缩伪影和训练速度慢等问题。针对上述问题,提出了分布式策略下的解码端增强图像压缩网络,用于减少重建图像压缩伪影和提高训练速度。一方面,原有的信息聚合子网不能有效利用超先验解码器的输出信息,不可避免地会在重建图像中产生压缩伪影。因此,使用解码端增强模块预测重建图像中的高频分量,减少压缩伪影。随后,为了进一步提高网络的非线性能力,引入了特征增强模块,进一步提高重建图像质量。另一方面,采用分布式训练解决单个节点训练速度较慢的问题,通过使用分布式训练有效缩短了训练时间。然而,分布式训练过程中梯度同步会产生大量通信开销,将梯度稀疏算法加入了分布式训练,每个节点按照概率将重要的梯度传递到主节点进行更新,进一步提高了训练速度。实验结果表明,分布式训练可以在保证重建图像质量的基础上加速训练。

关键词: 分布式训练, 解码端增强, 深度学习, 图像压缩

Abstract:

With the rapid development of multimedia,large-scale image data causes a great pressure on network bandwidth and storage.Presently,deep learning-based image compression methods still have problems such as compression artifacts in the reconstructed image and a slow training speed.To address the above problems,we propose a decoder-side enhanced image compression network under distributed strategy to reduce the reconstructed image compression artifacts and improve the training speed.On the one hand,the original information aggregation subnetwork cannot effectively utilize the output information of the hyperpriori decoder,which inevitably generates compression artifacts in the reconstructed image and negatively affects the visual effect of the reconstructed image.Therefore,we use the decoder-side enhancement module to predict the high-frequency components in the reconstructed image and reduce the compression artifacts.Subsequently,in order to further improve the nonlinear capability of the network,a feature enhancement module is introduced to further improve the reconstructed image quality.On the other hand,distributed training is introduced in this paper to solve the problem of slow training of traditional single node networks,and the training time is effectively shortened by using distributed training.However,the gradient synchronization during distributed training generates a large amount of communication overhead,so we add the gradient sparse algorithm to distributed training,and each node passes the important gradient to the master node for updating according to the probability,which further improves the training speed.Experimental results show that distributed training can accelerate training on the basis of ensuring the quality of the reconstructed image.

Key words: distributed training, decode-side enhancement, deep learning, image compression

中图分类号:

TP391.4

张静, 吴慧雪, 张少博, 李云松. 分布式策略下的解码端增强图像压缩网络[J]. 西安电子科技大学学报, 2025, 52(1): 1-13.

ZHANG Jing, WU Huixue, ZHANG Shaobo, LI Yunsong. Decoder-side enhanced image compression network under distributed strategy[J]. Journal of Xidian University, 2025, 52(1): 1-13.

图/表 12

图1

图2

图3

图4

图5

图6

表1

表2

图7

图8

图9

表3

参考文献 31

[1]	WALLACE G K. The JPEG Still Picture Compression Standard[J]. IEEE Transactions on Consumer Electronics, 1992, 38(1):xviii-xxxiv.
[2]	SKODRAS A, CHRISTOPOULOS C, EBRAHIMI T. The JPEG 2000 Still Image Compression Standard[J]. IEEE Signal Processing Magazine, 2001, 18(5):36-58.
[3]	LAINEMA J, BOSSEN F, HAN W J, et al. Intra Coding of The Hevc Standard[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2012, 22(12):1792-1801.
[4]	GAO H, ESENLIK S, ALSHINA E, et al. Geometric Partitioning Mode in Versatile Video Coding:Algorithm Review and Analysis[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2020, 31(9):3603-3617.
[5]	ZHANG J, WANG H, WANG Y, et al. Deep Network Based on Up and Down Blocks Using Wavelet Transform and Successive Multi-Scale Spatial Attention for Cloud Detection[J]. Remote Sensing of Environment, 2021, 261:112483.1-18.
[6]	ZHANG J, SHI X L, WU J, et al. Cloud Detection Method based on Spatial-Spectral Features and Encoder-Decoder Feature Fusion[J]. IEEE Transactions on Geoscience and Remote Sensing, 2023, 61:5407915.1-11.
[7]	ZHANG J, SHI X L, ZHENG C Y, et al. MRPFA-Net for Shadow Detection in Remote-Sensing Images[J]. IEEE Transactions on Geoscience and Remote Sensing, 2023, 61:5514011.1-11.
[8]	ZHANG J, SONG L N, WANG Y C, et al. Attention Mechanism with Spatial Spectrum Dense Connection and Context Dynamic Convolution for Cloud Detection[J]. IEEE Transactions on Geoscience and Remote Sensing, 2023, 61:5408211.1-11.
[9]	ZHANG J, ZHENG R J, CHEN X, et al. Spectral Correlation and Spatial High-Low Frequency Information of Hyperspectral Image Super-Resolution Network[J]. Remote Sensing, 2023, 15(9):2472.1-28.
[10]	ZHANG J, ZHENG R J, WAN Z K, et al. Hyperspectral Image Super-Resolution Based on Feature Diversity Extraction[J]. Remote Sensing, 2024, 16(3):436.1-24.
[11]	ZHANG J, HONG Z L, et al. Few-Shot Object Detection for Remote Sensing Imagery Using Segmentation Assistance and Triplet Head[J]. Remote Sensing, 2024, 16(19):3630.1-21.
[12]	王娟, 刘子杉, 武明虎, 等. 融合超分辨率重建技术的多尺度目标检测算法[J]. 西安电子科技大学学报, 2023, 50(3):122-131.
	WANG Juan, LIU Zishan, WU Minghu, et al. Multi-Scale Object Detection Algorithm Combined with Super-Resolution Reconstruction Technology[J]. Journal of Xidian University, 2023, 50(3):122-131.
[13]	刘晓雯, 郭继昌, 郑司达. 采用渐进式网络的弱监督显著性目标检测算法[J]. 西安电子科技大学学报, 2023, 50(1):48-57.
	LIU Xiaowen, GUO Jichang, ZHENG Sida. Weakly-Supervised Salient Object Detection with the Multi-Scale Progressive Network[J]. Journal of Xidian University, 2023, 50(1):48-57.
[14]	DONG C, DENG Y, LOY C C, et al. Compression Artifacts Reduction by ADeep Convolutional Network[C]//Proceedings of the IEEE International Conference on Computer Vision. Piscataway:IEEE, 2015:576-584.
[15]	TODERICI G, VINCENT D, JOHNSTON N, et al. Full Resolution Image Compression with Recurrent Neural Networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE, 2017:5306-5314.
[16]	LYU H, SHA N, QIN S, et al. Advancesin Neural Information Processing Systems[J]. Advances in Neural Information Processing Systems, 2019, 32:1-11.
[17]	SZEGEDY C, IOFFE S, VANHOUCKE V, et al. Inception-V4,Inception-Resnet and The Impact of Residual Connections on Learning[C]// Proceedings of the AAAI Conference on Artificial Intelligence. Palo Alto: AAAI, 2017:4278-4284.
[18]	DENG J, DONG W, SOCHER R, et al. Imagenet:A Large-Scale Hierarchical Image Database[C]//2009 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE, 2009:248-255.
[19]	LEE K. Introducing Big Basin:Our Next-Generation Ai Hardware(2017)[EB/OL].[2019-10-20]. https://code.facebook.com/posts/1835166200089399/introducing-big-basin.
[20]	JIA X, SONG S, HE W, et al. Highly Scalable Deep Learning Training System with Mixed-Precision:Training Imagenet in Four Minutes(2018)[J/OL].[2023-12-01]. https://arxiv.org/abs/1807.11205.
[21]	JIN Z, IQBAL M Z, ZOU W, et al. Dual-Stream Multi-Path Recursive Residual Network for JPEG Image Compression Artifacts Reduction[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2021, 31(2):467-479.
[22]	LIN M H, YEH C H, LIN C H, et al. Deep Multi-Scale Residual Learning-based Blocking Artifacts Reduction for Compressed Images[C]//Proceedings of the IEEE International Conference on Artificial Intelligence Circuits and Systems. Piscataway:IEEE, 2019:18-19.
[23]	ZHANG J, ZHANG S, WANG H, et al. Image Compression Network Structure Based on Multiscale Region of Interest Attention Network[J]. Remote Sensing, 2023, 15(2):522.
[24]	LI J, FANG F, MEI K, et al. Multi-Scale Residual Network for Image Super-resolution[C]//Proceedings of the European Conference on Computer Vision(ECCV). Heidelberg:Springer, 2018:517-532.
[25]	LIN Y, HAN S, MAO H, et al. Deep Gradient Compression:Reducing the Communication Bandwidth for Distributed Training(2017)[J/OL].[2023-01-05]. https://arxiv.org/abs/1712.01887.
[26]	KINGMA D P, BA J. Adam:A Method for Stochastic Optimization(2014)[J/OL].[2020-02-21]. https://arxiv.org/abs/1412.6980.
[27]	BALLÉ J, MINNEN D, SINGH S, et al. Variational Image Compression with AScale Hyperprior(2018)[J/OL].[2021-10-04]. https://arxiv.org/abs/1802.01436.
[28]	CHENG Z, SUN H, TAKEUCHI M, et al. Learned Image Compression with Discretized Gaussian Mixture Likelihoods and Attention Modules[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE, 2020:7939-7948.
[29]	HU Y, YANG W, LIU J. Coarse-O-Fine Hyper-Prior Modeling for Learned Image Compression[C]// Proceedings of the AAAI Conference on Artificial Intelligence. Palo Alto: AAAI, 2020:11013-11020.
[30]	HE D, ZHENG Y, SUN B, et al. Checkerboard Context Model for Efficient Learned Image Compression[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE, 2021:14771-14780.
[31]	XIANG S, LIANG Q. Remote Sensing Image Compression Based on High-Frequency and Low-Frequency Components[J]. IEEE Transactions on Geoscience and Remote Sensing, 2024, 62:1-15.

分布式训练算法	2张GPU	4张GPU	6张GPU	8张GPU
无梯度稀疏	396.27	260.43	177.23	137.38
ProbComp-LPAC	372.64	234.41	151.78	110.51

分布式训练算法	2张GPU	4张GPU	6张GPU	8张GPU
无梯度稀疏	0.098 31	0.098 29	0.098 35	0.098 42
ProbComp-LPAC	0.098 33	0.098 37	0.098 32	0.098 37

压缩算法	bpp	PSNR/dB	MS-SSIM/dB	PSNR_R/dB	MS-SSIM_R/dB
Baseline	0.305	31.31	16.29	35.17	18.19
	0.463	32.86	18.39	36.71	20.24
	1.199	37.01	23.74	40.81	25.59
Baseline+FNM	0.299	31.38	16.38	35.21	18.21
	0.457	32.98	18.45	36.85	20.45
	1.191	37.09	23.77	40.88	25.64
Baseline+DEM	0.295	31.40	16.39	35.29	18.23
	0.460	33.08	18.59	36.93	20.48
	1.184	37.10	23.81	40.91	25.71
文中方法	0.296	31.46	16.45	35.30	18.25
	0.456	33.11	18.59	36.93	20.48
	1.168	37.16	23.94	40.94	25.70
文中方法*	0.291	31.44	16.41	35.32	18.25
	0.455	33.13	18.69	36.99	20.51
	1.171	37.14	23.89	40.95	25.78