Survey on Computer Vision Techniques for Internet-of-Things Devices
Not on the map. Why?
Abstract (arXiv)
Deep neural networks (DNNs) are state-of-the-art techniques for solving most computer vision problems. DNNs require billions of parameters and operations to achieve state-of-the-art results. This requirement makes DNNs extremely compute, memory, and energy-hungry, and consequently difficult to deploy on small battery-powered Internet-of-Things (IoT) devices with limited computing resources. Deployment of DNNs on Internet-of-Things devices, such as traffic cameras, can improve public safety by enabling applications such as automatic accident detection and emergency response.Through this paper, we survey the recent advances in low-power and energy-efficient DNN implementations that improve the deployability of DNNs without significantly sacrificing accuracy. In general, these techniques either reduce the memory requirements, the number of arithmetic operations, or both. The techniques can be divided into three major categories: neural network compression, network architecture search and design, and compiler and graph optimizations. In this paper, we survey both low-power techniques for both convolutional and transformer DNNs, and summarize the advantages, disadvantages, and open research problems.
Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 29 of its 36 references that are in the map. First posted on arXiv 2 Aug 2023.
Builds on
29 of 36 references.
References · 29
- 2015Deep Residual Learning for Image Recognitioncited by 51,809
- 2014Very Deep Convolutional Networks for Large-Scale Image Recognitioncited by 15,267
- 2018MobileNetV2: Inverted Residuals and Linear Bottleneckscited by 4,773
- 2019ALBERT: A Lite BERT for Self-supervised Learning of Language Representationscited by 2,772
- 2015Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Codingcited by 2,291
- 2019Searching for MobileNetV3cited by 1,984
- 2016SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model sizecited by 1,054
- 2019EfficientDet: Scalable and Efficient Object Detectioncited by 1,018
- 2018MnasNet: Platform-Aware Neural Architecture Search for Mobilecited by 937
- 2018Regularized Evolution for Image Classifier Architecture Searchcited by 902
- 2016XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networkscited by 829
- 2018ProxylessNAS: Direct Neural Architecture Search on Target Task and Hardwarecited by 666
17 more
- 2018Object Detection with Deep Learning: A Reviewcited by 524
- 2018Quantizing deep convolutional networks for efficient inference: A whitepapercited by 507
- 2018FBNet: Hardware-Aware Efficient ConvNet Design via Differentiable Neural Architecture Searchcited by 480
- 2014Speeding up Convolutional Neural Networks with Low Rank Expansionscited by 283
- 2021Post-Training Quantization for Vision Transformercited by 274
- 2021I-BERT: Integer-only BERT Quantizationcited by 250
- 2015Quantized Convolutional Neural Networks for Mobile Devicescited by 212
- 2014Speeding-up Convolutional Neural Networks Using Fine-tuned CP-Decompositioncited by 199
- 2015Convolutional neural networks with low-rank regularizationcited by 102
- 2017Compression-aware Training of Deep Networkscited by 46
- 2018Recent progress in semantic image segmentationcited by 39
- 2020TRP: Trained Rank Pruning for Efficient Deep Neural Networkscited by 32
- 2022Neural Architecture Search on Efficient Transformers and Beyondcited by 17
- 2023Operator Fusion in XLA: Analysis and Evaluationcited by 8
- 2018NullaNet: Training Deep Neural Networks for Reduced-Memory-Access Inferencecited by 3
- 2019Low Rank Factorization for Compact Multi-Head Self-Attentioncited by 3
- 2022Irrelevant Pixels are Everywhere: Find and Exclude Them for More Efficient Computer Visioncited by 3
Built on by
Not cited yet.