params.json

{
  "name": "Squeezenet-residual",
  "tagline": "residual-SqueezeNet",
  "body": "# SqueezeNet-Residual\r\n\r\nThe repo contains the residual-SqueezeNet, which is obtained by adding bypass layer to SqueezeNet_v1.0. Residual-SqueezeNet improves the top-1 accuracy of SqueezeNet by 2.9% on ImageNet without changing the model size(only 4.8MB).\r\n\r\n# Related repo and paper\r\n[SqueezeNet](https://github.com/DeepScale/SqueezeNet)\r\n\r\n[SqueezeNet-Deep-Compression](https://github.com/songhan/SqueezeNet-Deep-Compression)\r\n\r\n[SqueezeNet-Generator](https://github.com/songhan/SqueezeNet-Generator)\r\n\r\n[SqueezeNet-DSD-Training](https://github.com/songhan/SqueezeNet-DSD-Training)\r\n\r\n[SqueezeNet-Residual](https://github.com/songhan/SqueezeNet-Residual)\r\n\r\n\r\n\r\nIf you find residual-SqueezeNet useful in your research, please consider citing the paper:\r\n\r\n    @article{SqueezeNet,\r\n      title={SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and< 0.5MB model size},\r\n      author={Iandola, Forrest N and Han, Song and Moskewicz, Matthew W and Ashraf, Khalid and Dally, William J and Keutzer, Kurt},\r\n      journal={arXiv preprint arXiv:1602.07360},\r\n      year={2016}\r\n    }\r\n  \r\n\r\n# Usage\r\n\r\n    $CAFFE_ROOT/build/tools/caffe test --model=trainval.prototxt --weights=SqueezeNet_residual_top1_0.6038_top5_0.8250.caffemodel --iterations=1000 --gpu 0\r\n\r\n# Result\r\n      \r\n    I0422 14:07:39.810755 32299 caffe.cpp:293] accuracy_top1 = 0.603759\r\n    I0422 14:07:39.810775 32299 caffe.cpp:293] accuracy_top5 = 0.824981\r\n    I0422 14:07:39.810792 32299 caffe.cpp:293] loss = 1.76711 (* 1 = 1.76711 loss) \r\n    \r\n# Architecture of the residual SqueezeNet\r\n<br>\r\n<img src=\"figure/architecture2.jpg\"  height=\"600px\" align=\"middle\" />\r\n\r\n\r\nThe building block:\r\n\r\n\r\n<img src=\"figure/type2.jpg\"  height=\"250px\" align=\"middle\"/>\r\n\r\n\r\n\r\n\r\n",
  "note": "Don't delete this file! It's used internally to help with page regeneration."
}