Feature-Align Network with Knowledge Distillation for Efficient Denoising

Lucas, Young,; Reda, Fitsum A.; Ranjan, Rakesh; Morton, Jon; Hu, Jun; Ling, Yazhu; Xiang, Xiaoyu; Liu, David; Chandra, Vikas

doi:10.1109/wacvw54805.2022.00078

Cited by 8 publications

(2 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Denoising is another important workload for AR/VR applications since image restoration is needed for a seamless and enjoyable user experience and denoising also helps improve detection rates given limited sensor resolution. 9 Increasing from 1 to 2 MB significantly reduces shared memory accesses in Figure 7 due to the relatively small size of the model and we see diminishing returns increasing the activation memory beyond 2 MB. Our simulator estimates $1.35Â total energy savings and 1.4-2.8Â total latency savings going from 2-D to 3-D.…”

Section: Denoisingmentioning

confidence: 66%

Three-Dimensional Stacked Neural Network Accelerator Architectures for AR/VR Applications

et al. 2022

View full text Add to dashboard Cite

Three-dimensional integration offers architectural and performance benefits for scaling augmented/virtual reality (AR/VR) models on highly resource-constrained edge devices. Two-dimensional off-chip memory interfaces are too prohibitively energy intensive and bandwidth (BW) limited for AR/VR devices. To solve this, we propose using advanced 3-D stacking technology for high-density vertical integration to local memory and compute, increasing memory capacity within the same footprint at iso-BW with improvements in energy and latency. We evaluate 3-D architectures for a prototype AR/VR accelerator to demonstrate up to 3.9Â latency reduction and 1.6Â lower energy compared to a 2-D configuration within a smaller/ similar footprint. Additionally, we show the feasibility of deploying higher resolution AR/VR models by stacking multiple tiers of memory, providing a pathway to break the footprint constraints of 2-D architectures. The use of high-density 3-D interconnects allows us to demonstrate localized benefits at the accelerator-level compared with standard system-on-chip memory disaggregation techniques/architectures.

show abstract

Section: Denoisingmentioning

confidence: 66%

Three-Dimensional Stacked Neural Network Accelerator Architectures for AR/VR Applications

et al. 2022

View full text Add to dashboard Cite

show abstract

“…Firstly, considering noise, a multi-scale encoder layer is proposed to extract high and low-frequency features from the input signals. Secondly, to deal with misalignment problems, following [53], a feature-alignment module is introduced in place of the deformable convolutional network [7], leading lower computational cost. Finally, a Progressive Dilated U-shape Block (PDUB) is proposed, composed of multiple tiny u-form structures to progressively restore features, achieving a good balance between PSNR-µ and GMACs in both Track 1 and Track 2.…”

Section: Antins CVmentioning

confidence: 99%

NTIRE 2021 Challenge on High Dynamic Range Imaging: Dataset, Methods and Results

Pérez-Pellitero

Catley-Chandar

Leonardis

et al. 2021

2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

View full text Add to dashboard Cite

This paper reviews the first challenge on high-dynamic range (HDR) imaging that was part of the New Trends in Image Restoration and Enhancement (NTIRE) workshop, held in conjunction with CVPR 2021. This manuscript focuses on the newly introduced dataset, the proposed methods and their results. The challenge aims at estimating a HDR image from one or multiple respective low-dynamic range (LDR) observations, which might suffer from underor over-exposed regions and different sources of noise. The challenge is composed by two tracks: In Track 1 only a single LDR image is provided as input, whereas in Track 2 three differently-exposed LDR images with inter-frame motion are available. In both tracks, the ultimate goal is to achieve the best objective HDR reconstruction in terms of PSNR with respect to a ground-truth image, evaluated both directly and with a canonical tonemapping operation.

show abstract

Toward Efficient Image Denoising: A Lightweight Network with Retargeting Supervision Driven Knowledge Distillation

Zou

Zhang

Wang

et al. 2022

Advances in Computer Graphics

View full text Add to dashboard Cite

Feature-Align Network with Knowledge Distillation for Efficient Denoising

Cited by 8 publications

References 36 publications

Three-Dimensional Stacked Neural Network Accelerator Architectures for AR/VR Applications

Three-Dimensional Stacked Neural Network Accelerator Architectures for AR/VR Applications

NTIRE 2021 Challenge on High Dynamic Range Imaging: Dataset, Methods and Results

Toward Efficient Image Denoising: A Lightweight Network with Retargeting Supervision Driven Knowledge Distillation

Contact Info

Product

Resources

About