Home / Current Issue / Paper 1722518
Real-Time Low-Power Super-Resolution on the Edge: A Comparative Study of Pruned GANs and Post-Training Quantization
Subject area: Science,Engineering and Technology · Area of research: Electrical and Electronic Engineering
DOI: https://doi.org/10.64388/IREV6I3-1722518
Abstract
Single-image super-resolution is increasingly demanded on mobile and embedded platforms, yet generative super-resolution networks are notoriously heavy. This paper studies how structured channel pruning combined with post-training quantization (PTQ) can deliver real-time 4x super-resolution within the power envelope of edge accelerators. We prune three GAN generators to 40% of their original channel width and quantize them to INT8 and INT4, then benchmark reconstruction fidelity, latency, and power on three representative devices. Our results show that pruning-then-PTQ preserves most perceptual quality while cutting latency by up to 47% and energy by roughly a third.
References
[1] [1] Wagner IS, Villanueva OT, Andric HW. Spatiotemporal Deep Networks for Motion Prediction in Cybersecurity. Knowledge-Based Systems. 2016;45(4):196-227. doi: 10.4491/j615660.2016.2738.
[2] [2] Suzuki AC, Pappas GT, Cardoso OG, Khedkar WC. Encoder-Decoder Networks for Feature Extraction in Digital Pathology. arXiv preprint arXiv:2223.52056. 2016. https://arxiv.org/abs/2223.52056 [Crossref]
[3] [3] Kaushik P, Jain M, Shah A. A Low Power Low Voltage CMOS Based Operational Transconductance Amplifier for Biomedical Application. 2018. https://ijsetr.com/uploads/136245IJSETR17012-283.pdf [Crossref]
[4] [4] Schneider NF, Salcedo EE, Rasmussen JO, Navarro EM. Federated Learning for Image Classification in Surveillance Systems. Information Sciences. 2015;59(1):209-229. https://doi.org/10.9422/w419640.2024.1052 [Crossref]
[5] [5] Kaushik P, Jain M. Design of low power CMOS low pass filter for biomedical application. J Electr Eng Technol (IJEET). 2018;9(5).
[6] [6] Villanueva HT, Marchetti PL, Chatterjee MC. Encoder-Decoder Networks: An Application to Signal Reconstruction. Pattern Recognition. 2024;18(4):63-98. https://doi.org/10.1225/t519651.2020.4721 [Crossref]
[7] [7] Kaushik P, Jain M. A Low Power SRAM Cell for High-Speed Applications Using 90nm Technology. Csjournals.Com. 10. https://www.csjournals.com/IJEE/PDF10-2/66.%20Puneet.pdf [Crossref]
[8] [8] Ghosh DI, Cardoso DM, Sharabi EB. Graph Neural Networks for Intrusion Detection in Edge Devices. IEEE Transactions on Neural Networks and Learning Systems. 2020. https://doi.org/10.1666/o610236.2022.6744 [Crossref]
[9] [9] Kaushik P, Jain M. A Low Power SRAM Cell for High-Speed Applications Using 90nm Technology. Csjournals.Com. 2018 Dec;10(2):6. https://www.csjournals.com/IJEE/PDF10-2/66.%20Puneet.pdf [Crossref]
[10] [10] Escobar NK, Saito DV. A Hybrid CNN-Transformer Model for Fraud Detection in IoT Networks. Artificial Intelligence in Medicine. 2018;26(1):249-255. doi: 10.4711/u753045.2024.3491.
[11] [11] Kaushik P, Jain M, Jain A. A Pixel-Based Digital Medical Images Protection Using Genetic Algorithm. International Journal of Electronics and Communication Engineering. 2018;11(1):31-37. ISSN 0974-2166.
[12] [12] Andersson LW, Zhang IF, Dominguez GC. Lightweight Convolutional Networks: An Application to Signal Reconstruction. Neurocomputing. 2015. https://doi.org/10.4914/k238768.2021.5429 [Crossref]
[13] [13] Kaushik P, Jain M, Shah A. A Low Power Low Voltage CMOS Based Operational Transconductance Amplifier for Biomedical Application. 2018.
[14] [14] Cardoso NC, Tanaka OD, Wang SC, Ortega WP. Ensemble Deep Learning for Lesion Segmentation in Medical Imaging. IEEE Transactions on Image Processing. 2018. https://doi.org/10.2684/e921436.2016.1030 [Crossref]
[15] [15] Chatterjee WA, Duarte GO, Zapata JM. Metric Learning for Pose Estimation in Smart Manufacturing. IEEE Transactions on Image Processing. 2015. https://doi.org/10.2179/f806330.2021.4636 [Crossref]
[16] [16] Kaushik P, Jain M. A Low Power SRAM Cell for High-Speed Applications Using 90nm Technology. Csjournals.Com. 2018 Dec;10(2):6. https://www.csjournals.com/IJEE/PDF10-2/66.%20Puneet.pdf [Crossref]
[17] [17] Oliveira KL, Yamamoto BA, Andersson EI. Federated Learning: An Application to Image Super-Resolution. IEEE Transactions on Neural Networks and Learning Systems. 2020;57(5):246-274. doi: 10.3973/g554998.2023.1371.
[18] [18] Duarte RB, Silva CP, Petrov WW. U-Shaped Convolutional Networks for Object Detection in Edge Devices. Expert Systems with Applications. 2022;46(6):114-135. https://doi.org/10.6430/r770629.2023.9193 [Crossref]
[19] [19] Jain M, Shah A. Machine Learning with Convolutional Neural Networks (CNNs) in Seismology for Earthquake Prediction. Iconic Research and Engineering Journals. 2022;5(8):389-398. https://www.irejournals.com/paper-details/1707057 [Crossref]
[20] [20] Brandt WK, Marchetti VI, Rossi OK, Takahashi KT. Deep Residual Learning for Scene Understanding in Neuroimaging. IEEE Transactions on Neural Networks and Learning Systems. 2019. https://doi.org/10.1262/s807910.2020.6694 [Crossref]
[21] [21] Jain M, Srihari A. Comparison of CAD detection of mammogram with SVM and CNN. IRE Journals. 2021;8(6):63-75. https://www.irejournals.com/formatedpaper/1706647.pdf [Crossref]
[22] [22] Ghosh ME, Larsson OT, Marchetti CN. Generative Adversarial Networks for Signal Reconstruction in Healthcare Systems. Pattern Recognition. 2022;48(7):62-80. https://doi.org/10.5085/d372609.2018.4047 [Crossref]
[23] [23] Bianchi SC, Mercado FN. U-Shaped Convolutional Networks for Pose Estimation in Clinical Diagnostics. Knowledge-Based Systems. 2018;21(11):330-341. https://doi.org/10.6515/e624896.2016.9065 [Crossref]
[24] [24] Saito BB, Ivanov FT, Takahashi HM. U-Shaped Convolutional Networks for Lesion Segmentation in Financial Systems. Multimedia Tools and Applications. 2021;26(1):188-207. https://doi.org/10.7387/d682026.2016.8368 [Crossref]
[25] [25] Serrano AL, Leung SD, Garofalo CH. Federated Learning: An Application to Signal Reconstruction. Pattern Recognition. 2015. https://doi.org/10.4542/n124969.2015.7719 [Crossref]
[26] [26] Ferreira VO, Grigoriev HK, Khedkar FA. A Comparative Study of U-Shaped Convolutional Networks for Lesion Segmentation. Multimedia Tools and Applications. 2018. https://doi.org/10.4229/i750703.2023.5651 [Crossref]
[27] [27] Kaushik P, Jain M. A Low Power SRAM Cell for High Speed Applications Using 90nm Technology. Csjournals.Com. 10. https://www.csjournals.com/IJEE/PDF10-2/66.%20Puneet.pdf [Crossref]
[28] [28] Haddad FW, Jimenez BT, Novak LO. A Comparative Study of Capsule Networks for Image Denoising. Scientific Reports. 2016. https://doi.org/10.7405/r888860.2015.7753 [Crossref]
[29] [29] Khedkar TI, Delgado BV. Transfer Learning for Object Detection in Surveillance Systems. IEEE Transactions on Image Processing. 2023;22(1):188-201. https://doi.org/10.6696/a315873.2024.7201 [Crossref]
[30] [30] Kaushik P, Jain M, Jain A. A pixel-based digital medical images protection using genetic algorithm. Int J Electron Commun Eng. 2018;11:31-37.
[31] [31] Jain M, Shah A. Convolutional neural networks for real-time object detection with raspberry Pi. 2021. https://wjaets.com/sites/default/files/WJAETS-2021-0067.pdf. https://doi.org/10.30574/wjaets.2021.4.1.0067 [Crossref]
[32] [32] Brandt JJ, Khedkar CV, Krause TT, Thakur GV. U-Shaped Convolutional Networks: An Application to Feature Extraction. Applied Soft Computing. 2023;6(12):85-122. https://doi.org/10.5487/e219455.2019.6016 [Crossref]
[33] [33] Jain M, Shah A. A multi-modal CNN framework for integrating medical imaging for COVID-19 Diagnosis. World Journal of Advanced Research and Reviews. 2020;8(3):475-493. https://doi.org/10.30574/wjarr.2020.8.3.0418 [Crossref]
[34] [34] Escobar VI, Lindberg AA. Federated Learning for Lesion Segmentation in Cybersecurity. Artificial Intelligence in Medicine. 2020. https://doi.org/10.9582/r609267.2024.6756 [Crossref]
[35] [35] Reinholt SM, Mahmood CE, Larsson GK. Bayesian Deep Learning for Medical Image Synthesis in Clinical Diagnostics. Medical Image Analysis. 2015;15(12):384-419. https://doi.org/10.4480/k222730.2022.4340 [Crossref]
[36] [36] Novak NT, Nascimento NM. Transfer Learning for Image Super-Resolution in Neuroimaging. Computers in Biology and Medicine. 2023;66(12):169-195. https://doi.org/10.8357/h128322.2018.2910 [Crossref]
[37] [37] Kaushik P. STUDY AND ANALYSIS OF IMAGE ENCRYPTION ALGORITHM BASED ON ARNOLD TRANSFORMATION. INTERNATIONAL JOURNAL of COMPUTER ENGINEERING and TECHNOLOGY (IJCET). 2018;9(5):59-63. https://iaeme.com/Home/article_id/IJCET_09_05_008 [Crossref]
[38] [38] Cardoso TI, Nascimento BN, Haddad FW, Petrov IG. A Comparative Study of Deep Residual Learning for Image Segmentation. IEEE Journal of Biomedical and Health Informatics. 2023;49(6):144-170. https://doi.org/10.9204/b190889.2017.8242 [Crossref]
[39] [39] Kaushik P, Jain M. Design of low power CMOS low pass filter for biomedical application. International Journal of Electrical Engineering & Technology (IJEET). 2018;9(5).
[40] [40] Yilmaz GV, Kallas TT, Petrov PV. Knowledge Distillation: An Application to Semantic Segmentation. Pattern Recognition. 2016. https://doi.org/10.6576/f777964.2016.7312 [Crossref]
[41] [41] Sokolov EA, Silva PA, Pappas TF. Capsule Networks for Anomaly Detection in Radiology. arXiv preprint arXiv:2409.37983. 2020. https://arxiv.org/abs/2409.37983 [Crossref]
[42] [42] Feng GN, Cardoso BC, Fischer LR, Sandberg KF. Spatiotemporal Deep Networks: An Application to Semantic Segmentation. IEEE Transactions on Neural Networks and Learning Systems. 2021;12(3):82-111. https://doi.org/10.7917/y203899.2022.3307 [Crossref]
[43] [43] Wang SA, Kallas SM, Wagner VL. Contrastive Representation Learning for Disease Classification in Edge Devices. Expert Systems with Applications. 2021. https://doi.org/10.7442/a227062.2017.6092 [Crossref]
[44] [44] Jain M, Kaushik P, Shah A. Comparison of VGG16 and VGG19 Convolutional Neural Network (CNN) Layers on MRI Brain Tumor Detection. International Journal of Trend in Scientific Research and Development (IJTSRD). 2016 Dec;1(1):275-280. ISSN 2456-6470. https://www.ijtsrd.com/papers/ijtsrd3542.pdf [Crossref]
[45] [45] Xu DB, Silva MV, Costa PP. Recurrent Neural Networks for Signal Reconstruction in Neuroimaging. Expert Systems with Applications. 2022;34(6):267-287. doi: 10.8630/z463790.2022.1373.
[46] [46] Serrano IN, Verduzco GF, Mahmood SG. Generative Adversarial Networks: An Application to Anomaly Detection. arXiv preprint arXiv:2153.50606. 2017. https://arxiv.org/abs/2153.50606 [Crossref]
How to cite this paper
@article{1722518,
author = {Hiroaki Tan, Sofia Marchetti, Rahul Bansal, Kwame Osei},
title = {Real-Time Low-Power Super-Resolution on the Edge: A Comparative Study of Pruned GANs and Post-Training Quantization},
journal = {Iconic Research And Engineering Journals},
year = {2022},
volume = {6},
number = {3},
pages = {397-402},
issn = {2456-8880},
url = {https://www.irejournals.com/formatedpaper/1722518.pdf},
abstract = {Single-image super-resolution is increasingly demanded on mobile and embedded platforms, yet generative super-resolution networks are notoriously heavy. This paper studies how structured channel pruning combined with post-training quantization (PTQ) can deliver real-time 4x super-resolution within the power envelope of edge accelerators. We prune three GAN generators to 40% of their original channel width and quantize them to INT8 and INT4, then benchmark reconstruction fidelity, latency, and power on three representative devices. Our results show that pruning-then-PTQ preserves most perceptual quality while cutting latency by up to 47% and energy by roughly a third.},
month = {September},
doi = {https://doi.org/10.64388/IREV6I3-1722518}
}