MixNet: [Arxiv]
by Wenhai Wang, Xiang Li, Jian Yang, Tong Lu
IMAGINE Lab@National Key Laboratory for Novel Software Technology, Nanjing University.
DeepInsight@PCALab, Nanjing University of Science and Technology.
- Install PyTorch v0.2.0
- Clone recursively
git clone --recursive https://github.com/DeepInsight-PCALab/MixNet.git
- Download the ImageNet dataset and move validation images to labeled subfolders
- To do this, you can use the following script: https://raw.githubusercontent.com/soumith/imagenetloader.torch/master/valprep.sh
CUDA_VISIBLE_DEVICES=0 python cifar.py --dataset cifar10 --depth 100 --k1 12 --k2 12 --train-batch 64 --epochs 300 --schedule 150 225 --wd 1e-4 --gamma 0.1 --checkpoint checkpoints/cifar10/mixnet-100/
CUDA_VISIBLE_DEVICES=0,1,2,3 python imagenet.py -d ../imagenet/ -j 4 --arch mixnet105 --train-batch 200 --checkpoint checkpoints/imagenet/mixnet-105/
CUDA_VISIBLE_DEVICES=0 python imagenet.py -d ../imagenet/ -j 4 --arch mixnet105 --test-batch 20 --pretrained pretrained/mixnet105.pth.tar --evaluate
| Model | Parameters | CIFAR-10 | CIFAR-100 |
|---|---|---|---|
| MixNet-100 (k1 = 12, k2 = 12) | 1.5M | 4.19 | 21.12 |
| MixNet-250 (k1 = 24, k2 = 24) | 29.0M | 3.32 | 17.06 |
| MixNet-190 (k1 = 40, k2 = 40) | 48.5M | 3.13 | 16.96 |
| Method | Parameters | Top-1 error | Pretrained model |
|---|---|---|---|
| MixNet-105 (k1 = 32, k2 = 32) | 11.16M | 23.3 | baidu, onedrive |
| MixNet-121 (k1 = 40, k2 = 40) | 21.86M | 21.9 | baidu, onedrive |
| MixNet-141 (k1 = 48, k2 = 48) | 41.07M | 20.4 | baidu, onedrive |
@inproceedings{wang2018mixed,
title={Mixed link networks},
author={Wang, Wenhai and Li, Xiang and Lu, Tong and Yang, Jian},
booktitle={Proceedings of the 27th International Joint Conference on Artificial Intelligence},
pages={2819--2825},
year={2018},
organization={AAAI Press}
}