SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size
Citations
14,406 citations
Cites background from "SqueezeNet: AlexNet-level accuracy ..."
...Another small network is Squeezenet [12] which uses a bottleneck approach to design a very small network....
[...]
...It is also 4% better than Squeezenet [12] at about the same size and 22× less computation....
[...]
7,256 citations
6,222 citations
Additional excerpts
...As mobile phones become ubiquitous, it is also common to handcraft efficient mobile-size ConvNets, such as SqueezeNets (Iandola et al., 2016; Gholami et al., 2018), MobileNets (Howard et al., 2017; Sandler et al., 2018), and ShuffleNets (Zhang et al., 2018; Ma et al., 2018)....
[...]
5,709 citations
Additional excerpts
...For those detectors running on CPU platform, their backbone could be SqueezeNet [31], MobileNet [28, 66, 27, 74], or ShuffleNet [97, 53]....
[...]
4,503 citations
References
123,388 citations
73,978 citations
55,235 citations
"SqueezeNet: AlexNet-level accuracy ..." refers background in this paper
...The VGG [26] architectures have 3x3 spatial resolution in most layers’ filters; GoogLeNet [32] and Network-in-Network (NiN) [22] have 1x1 filters in some layers....
[...]
...AlexNet and VGG [19]), but it is also able to compress the already compact, fully convolutional SqueezeNet architecture....
[...]
...In addition, these results demonstrate that Deep Compression [10] not only works well on CNN architectures with many parameters (e.g. AlexNet and VGG), but it is also able to compress the already compact, fully convolutional SqueezeNet architecture....
[...]
...The early work by LeCun et al. [21] uses 5x5xChannels2 filters, and the recent VGG [26] architectures extensively use 3x3 filters....
[...]
...Simoyan and Zisserman proposed the VGG [26] family of CNNs with 12 to 19 layers and reported that deeper networks produce higher accuracy on the ImageNet-1k dataset [4]....
[...]
49,914 citations
49,639 citations
"SqueezeNet: AlexNet-level accuracy ..." refers methods in this paper
...However, it has become common practice to apply ImageNet-trained CNN representations to a variety of applications such as fine-grained object recognition [34, 6], logo identification in images [17], and generating sentences about images [7]....
[...]
...In each of the CNN model compression papers reviewed in Section 2.1, the goal was to compress an AlexNet [20] model that was trained to classify images using the ImageNet [4] (ILSVRC 2012) dataset....
[...]
...ImageNet-trained CNNs have also been applied to a number of applications pertaining to autonomous driving, including pedestrian and vehicle detection in images [16, 8] and videos [3], as well as segmenting the shape of the road [1]....
[...]
...From this figure, we learn that increasing SR beyond 0.125 can further increase ImageNet top-5 accuracy from 80.3% (i.e. AlexNet-level) with a 4.8MB model to 86.0% with a 19MB model....
[...]
...We see in Figure 3(b) that the top-5 accuracy plateaus at 85.6% using 50% 3x3 filters, and further increasing the percentage of 3x3 filters leads to a larger model size but provides no improvement in accuracy on ImageNet....
[...]