An improved version of SqueezeNet networks https://github.com/DeepScale/SqueezeNet
Simplified SqueezeNet is SqueezeNet network with removing Fire/Expand layer of reception field 1 and the concatenation layer. The suggested modification accelerated the training of SqueezeNet by more than 4 times.
To understand the differences between the two networks, refer the the below links.
This link visualize the SqueezeNet network: http://ethereon.github.io/netscope/#/gist/d2bf1b838c4c07a465afbdcfac0577f9
This link visualize the Simplified SqueezeNet network: http://ethereon.github.io/netscope/#/gist/fd7f749c3cfe69a6859cc56225c10c0f
The above directories contain the source files for training SqueezeNet and Simplified SqueezeNet on two large scale image datasets.
Stay tuned for the technical report (coming soon)
For any information , please contact us at [email protected]
If you find the network useful in your research, please consider citing:
@article{SqueezeNet,
Author = {Hussein Al-barazanchi},
Title = {Simplified SqueezeNet},
Journal = {arXiv},
Year = {2017}
}