Effnet: An Efficient Structure for Convolutional Neural Networks

Freeman, Ido; Roese-Koerner, Lutz; Kummert, Anton

doi:10.1109/icip.2018.8451339

Cited by 101 publications

(78 citation statements)

References 18 publications

(44 reference statements)

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…There are a few possibilities for decreasing the computational demand in our approach. First of all, a lot of research has been done on creating efficient Convolutional Neural Network architectures for low latency predictions on mobile hardware [34,35]. We did not focus on efficiency and it should be relatively easy to decrease the computational load by optimizing the architecture.…”

Section: Conclusion and Discussionmentioning

confidence: 99%

Nonlinear wavefront reconstruction with convolutional neural networks for Fourier-based wavefront sensors

Landman¹,

Haffert²

2020

Opt. Express

View full text Add to dashboard Cite

Fourier-based wavefront sensors, such as the Pyramid Wavefront Sensor (PWFS), are the current preference for high contrast imaging due to their high sensitivity. However, these wavefront sensors have intrinsic nonlinearities that constrain the range where conventional linear reconstruction methods can be used to accurately estimate the incoming wavefront aberrations. We propose to use Convolutional Neural Networks (CNNs) for the nonlinear reconstruction of the wavefront sensor measurements. It is demonstrated that a CNN can be used to accurately reconstruct the nonlinearities in both simulations and a lab implementation. We show that solely using a CNN for the reconstruction leads to suboptimal closed loop performance under simulated atmospheric turbulence. However, it is demonstrated that using a CNN to estimate the nonlinear error term on top of a linear model results in an improved effective dynamic range of a simulated adaptive optics system. The larger effective dynamic range results in a higher Strehl ratio under conditions where the nonlinear error is relevant. This will allow the current and future generation of large astronomical telescopes to work in a wider range of atmospheric conditions and therefore reduce costly downtime of such facilities.

show abstract

Section: Conclusion and Discussionmentioning

confidence: 99%

Nonlinear wavefront reconstruction with convolutional neural networks for Fourier-based wavefront sensors

Landman¹,

Haffert²

2020

Opt. Express

View full text Add to dashboard Cite

show abstract

“…Currently, the design of computationally efficient CNNs is moving from manual tuning [17][18][19] towards automatic algorithms [20][21][22][23][24][25][26]. The incorporation of specific platform constraints to such approaches involves modeling how the network architecture relates with the optimization target.…”

Section: Related Workmentioning

confidence: 99%

PreVIous: A Methodology for Prediction of Visual Inference Performance on IoT Devices

Velasco-Montero

Fernández-Berni

Carmona-Galán

et al. 2020

IEEE Internet Things J.

View full text Add to dashboard Cite

This paper presents PreVIous, a methodology to predict the performance of convolutional neural networks (CNNs) in terms of throughput and energy consumption on vision-enabled devices for the Internet of Things. CNNs typically constitute a massive computational load for such devices, which are characterized by scarce hardware resources to be shared among multiple concurrent tasks. Therefore, it is critical to select the optimal CNN architecture for a particular hardware platform according to prescribed application requirements. However, the zoo of CNN models is already vast and rapidly growing. To facilitate a suitable selection, we introduce a prediction framework that allows to evaluate the performance of CNNs prior to their actual implementation. The proposed methodology is based on PreVIousNet, a neural network specifically designed to build accurate per-layer performance predictive models. PreVIousNet incorporates the most usual parameters found in state-of-the-art network architectures. The resulting predictive models for inference time and energy have been tested against comprehensive characterizations of seven well-known CNN models running on two different software frameworks and two different embedded platforms. To the best of our knowledge, this is the most extensive study in the literature concerning CNN performance prediction on low-power low-cost devices. The average deviation between predictions and real measurements is remarkably low, ranging from 3% to 10%. This means state-of-the-art modeling accuracy. As an additional asset, the fine-grained a priori analysis provided by PreVIous could also be exploited by neural architecture search engines.

show abstract

“…Following these observations, we define an image patch as hard-to-upscale based on the following criterion: (9) where the TV threshold TV thr is a tunable parameter whose value is automatically configured by MobiSR as discussed in Section 3.5.…”

Section: Difficulty Evaluation Unitmentioning

confidence: 99%

“…So far, substantial effort has been invested on developing network compression techniques, such as pruning [12,57], quantization [11], and knowledge distillation [18], for building efficient neural networks. In particular, a number of convolution approximations, such as low-rank tensor decomposition [47], have been successfully employed as a primary component in building fast and accurate discriminative vision models [9,19,58,61]. These techniques typically aim to express a convolution as a sequence of simpler tensor operations, reducing in this manner the storage and computation cost of the network.…”

Section: Introductionmentioning

confidence: 99%

MobiSR

Lee

Venieris

Dudziak

et al. 2019

The 25th Annual International Conference on Mobile Computing and Networking

View full text Add to dashboard Cite

In recent years, convolutional networks have demonstrated unprecedented performance in the image restoration task of super-resolution (SR). SR entails the upscaling of a single low-resolution image in order to meet application-specific image quality demands and plays a key role in mobile devices. To comply with privacy regulations and reduce the overhead of cloud computing, executing SR models locally on-device constitutes a key alternative approach. Nevertheless, the excessive compute and memory requirements of SR workloads pose a challenge in mapping SR networks on resource-constrained mobile platforms. This work presents MobiSR, a novel framework for performing efficient superresolution on-device. Given a target mobile platform, the proposed framework considers popular model compression techniques and traverses the design space to reach the highest performing trade-off between image quality and processing speed. At run time, a novel scheduler dispatches incoming image patches to the appropriate model-engine pair based on the patch's estimated upscaling difficulty in order to meet the required image quality with minimum processing latency. Quantitative evaluation shows that the proposed framework yields on-device SR designs that achieve an average speedup of 2.13× over highly-optimized parallel difficulty-unaware mappings and 4.79× over highly-optimized single compute engine implementations.

show abstract

Effnet: An Efficient Structure for Convolutional Neural Networks

Cited by 101 publications

References 18 publications

Nonlinear wavefront reconstruction with convolutional neural networks for Fourier-based wavefront sensors

Nonlinear wavefront reconstruction with convolutional neural networks for Fourier-based wavefront sensors

PreVIous: A Methodology for Prediction of Visual Inference Performance on IoT Devices

MobiSR

Contact Info

Product

Resources

About