SqueezeNet

SqueezeNet was originally released on February 22, 2016.{{Cite web|url=https://github.com/DeepScale/SqueezeNet|title=SqueezeNet|date=2016-02-22|website=GitHub|access-date=2018-05-12}} This original version of SqueezeNet was implemented on top of the Caffe deep learning software framework. Shortly thereafter, the open-source research community ported SqueezeNet to a number of other deep learning frameworks. On February 26, 2016, Eddie Bell released a port of SqueezeNet for the Chainer deep learning framework.{{Cite web|url=https://github.com/ejlb/squeezenet-chainer|title=An implementation of SqueezeNet in Chainer|last=Bell|first=Eddie|date=2016-02-26|website=GitHub|access-date=2018-05-12}} On March 2, 2016, Guo Haria released a port of SqueezeNet for the Apache MXNet framework.{{Cite web|url=https://github.com/hariag/SqueezeNet|title=SqueezeNet for MXNet|last=Haria|first=Guo|date=2016-03-02|website=GitHub|access-date=2018-05-12}} On June 3, 2016, Tammy Yang released a port of SqueezeNet for the Keras framework.{{Cite web|url=https://github.com/DT42/squeezenet_demo|title=SqueezeNet Keras Implementation|last=Yang|first=Tammy|date=2016-06-03|website=GitHub|access-date=2018-05-12}} In 2017, companies including Baidu, Xilinx, Imagination Technologies, and Synopsys demonstrated SqueezeNet running on low-power processing platforms such as smartphones, FPGAs, and custom processors.{{Cite news|url=https://www.theregister.co.uk/2017/09/26/baidu_mobile_deep_learning/|title=Baidu puts open source deep learning into smartphones|last=Chirgwin|first=Richard|date=2017-09-26|work=The Register|access-date=2018-04-07}}{{Cite news|url=https://www.electronicsweekly.com/news/products/software-products/neural-network-sdk-powervr-gpus-2018-01/|title=Neural network SDK for PowerVR GPUs|last=Bush|first=Steve|date=2018-01-25|work=Electronics Weekly|access-date=2018-04-07}}{{Cite news|url=https://www.eetimes.com/document.asp?doc_id=1331467&print=yes|title=Xilinx AI Engine Steers New Course|last=Yoshida|first=Junko|date=2017-03-13|work=EE Times|access-date=2018-05-13}}{{Cite news|url=https://www.engineerlive.com/content/deep-learning-computer-vision-algorithms-ported-processor-ip|title=Deep learning computer vision algorithms ported to processor IP|last=Boughton|first=Paul|date=2017-08-28|work=Engineer Live|access-date=2018-04-07}}

As of 2018, SqueezeNet ships "natively" as part of the source code of a number of deep learning frameworks such as PyTorch, Apache MXNet, and Apple CoreML. In addition, third party developers have created implementations of SqueezeNet that are compatible with frameworks such as TensorFlow. Below is a summary of frameworks that support SqueezeNet.

SqueezeNet was originally described in SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size.{{cite arXiv|eprint=1602.07360|title=SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size|last1=Iandola|first1=Forrest N|last2=Han|first2=Song|last3=Moskewicz|first3=Matthew W|last4=Ashraf|first4=Khalid|last5=Dally|first5=William J|last6=Keutzer|first6=Kurt|class=cs.CV|year=2016}} AlexNet is a deep neural network that has 240 MB of parameters, and SqueezeNet has just 5 MB of parameters. This small model size can more easily fit into computer memory and can more easily be transmitted over a computer network. However, it's important to note that SqueezeNet is not a "squeezed version of AlexNet." Rather, SqueezeNet is an entirely different DNN architecture than AlexNet.{{Cite web|url=http://www.shortscience.org/paper?bibtexKey=journals/corr/IandolaMAHDK16#unmesh|title=SqueezeNet|website=Short Science|access-date=2018-05-13}} What SqueezeNet and AlexNet have in common is that both of them achieve approximately the same level of accuracy when evaluated on the ImageNet image classification validation dataset.

Model compression (e.g. quantization and pruning of model parameters) can be applied to a deep neural network after it has been trained.{{Citation |last1=Han |first1=Song |title=Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Coding |date=2016-02-15 |arxiv=1510.00149 |last2=Mao |first2=Huizi |last3=Dally |first3=William J.}} In the SqueezeNet paper, the authors demonstrated that a model compression technique called Deep Compression can be applied to SqueezeNet to further reduce the size of the parameter file from 5 MB to 500 KB. Deep Compression has also been applied to other DNNs, such as AlexNet and VGG.{{Cite web|url=https://www.oreilly.com/ideas/compressing-and-regularizing-deep-neural-networks|title=Compressing and regularizing deep neural networks|last=Han|first=Song|date=2016-11-06|website=O'Reilly|access-date=2018-05-08}}

Some of the members of the original SqueezeNet team have continued to develop resource-efficient deep neural networks for a variety of applications. A few of these works are noted in the following table. As with the original SqueezeNet model, the open-source research community has ported and adapted these newer "squeeze"-family models for compatibility with multiple deep learning frameworks.

In addition, the open-source research community has extended SqueezeNet to other applications, including semantic segmentation of images and style transfer.{{Cite news|url=https://openreview.net/pdf?id=S1uHiFyyg|title=Speeding up Semantic Segmentation for Autonomous Driving|last=Treml|first=Michael|date=2016|work=NIPS MLITS Workshop|access-date=2019-07-21|display-authors=etal}}{{Cite web|url=https://github.com/lizeng614/SqueezeNet-Neural-Style-Pytorch|title=SqueezeNet Neural Style on PyTorch|last=Zeng|first=Li|date=2017-03-22|website=GitHub|access-date=2019-07-21}}{{Cite web|url=https://aspire.eecs.berkeley.edu/wiki/_media/eop/2017/bichen_talk_slides.pdf|title=The Impact of SqueezeNet|last1=Wu|first1=Bichen|last2=Keutzer|first2=Kurt|date=2017|website=UC Berkeley|access-date=2019-07-21}}

• Convolutional neural network
• MobileNet
• EfficientNet
• You Only Look Once
• Edge computing

{{Differentiable computing}}

Category:Deep learning software applications

Category:Artificial neural networks

Category:Object recognition and categorization

Category:Software using the BSD license