Code Implementation for Restricted Isometry Property(RIP) based Orthogonal Regularizers, proposed for Image Classification Task, for various State-of-art ResNet based architectures.
This repositry provides an introduction, implementation and result achieved in the paper: "Can We Gain More from Orthogonality Regularizations in Training Deep CNNs?", NeurIPS 2018 [pdf]
Orthogonal Network Weights are found to be a favorable property for training deep convolutional neural networks.Through this work, we look to find alternate and more effective ways to enforce orthogonality to deep CNNs. We develop novel orthogonality regularizations on training deep CNNs, utilizing various advanced analytical tools such as mutual coherence and restricted isometry property. These plug-and-play regularizations can be conveniently incorporated into training almost any CNN without extra hassle. We then benchmark their effects on state-of-the-art models: ResNet, WideResNet, and ResNeXt, on several most popular computer vision datasets: CIFAR-10, CIFAR-100, SVHN and ImageNet. We observe consistent performance gains after applying those proposed regularizations, in terms of both the final accuracies achieved, and faster and more stable convergences.
Figure 1. Validation Curve Achieved for differnet Regularizers Proposed
- Linux
- Pytorch 4.0
- Keras 2.2.4
- CUDA 9.1
- Cifar10 and Cifar100
- SVHN
- ImageNet
- ResNet
- Wide ResNet
- Pre Resnet
- ResNext
- Single Sided (SO)
- Double Sided (DSO)
- Mutual Coherence Based (MC)
- Restricted Isometry (SRIP) (Best Performing )
For CIFAR datasets,we choose Wide Resnet Arch. with a depth of 28 and Kernel width of 10,which gives the best results for comparable number parameters for any Wide-Resnet Model. To train on Cifar-10 using 2 gpu:
CUDA_VISIBLE_DEVICES=6,7 python train_n.py --ngpu 2
To train on Cifar-100 using 2 gpu:
CUDA_VISIBLE_DEVICES=6,7 python train_n.py --ngpu 2 --dataset cifar100
After train phase, you can check saved model in the runs
folder.
For SVHN datasets,we choose Wide Resnet Arch. with a depth of 16 and Kernel width of 8,which gives the best results for comparable number parameters for any Wide-Resnet Model.
CUDA_VISIBEL_DEVICES=0 python train.py --dataset svhn --model wideresnet --learning_rate 0.01 --epochs 160
Network | CIFAR-10 | CIFAR-100 | SVHN |
---|---|---|---|
WideResNet | 4.16 | 20.50 | 1.60 |
WideResNet + SRIP Reg | 3.60 | 18.19 | 1.52 |
We trained CIFAR10 and 100 Dataset for ResNet110 Model and achieved an improvement in terms of Test Accuracy, when compared to a model, which doesn't uses any form Regularization.The Code for this part has been written in Keras, and we have used the base code from official keras Repo: https://github.com/keras-team/keras/blob/master/examples/cifar10_resnet.py, for a bottleneck based architecture.
CUDA_VISIBLE_DEVICES=2 python resnet_cifar_new.py
Network | CIFAR-10 |
---|---|
ResNet110 | 7.11 |
WideResNet + SRIP Reg | 5.46 |
we trained the Imagenet Dataset for Resnet-34 Resnet 50 and Pre-Resnet 34 and achieved a better Top-5 accuracy when compared to contemporary results. Basic Code was taken from:Pytorch Official cite.
CUDA_VISIBLE_DEVICES=4,5,6,7 python train_n.py
Network | Imagenet | Regularizer |
---|---|---|
PreResnet 34 | 9.79 | NONE |
PreResNet 34 | 8. 85 | SRIP |
ResNet 34 | 9.84 | NONE |
ResNet 34 | 8.392 | SRIP |
Pre-trained Model are saved on the Google Drive.
- torch (@facebookresearch). (Original) Cifar and Imagenet
- [ResNet] (https://arxiv.org/pdf/1512.03385.pdf)
- [Pre ResNet] (https://arxiv.org/abs/1603.05027)
- [Wide Resnet] (https://arxiv.org/abs/1605.07146)
- [ResNext] (https://arxiv.org/abs/1611.05431)
If you find our code helpful in your resarch or work, please cite our paper.
@ARTICLE{2018arXiv181009102B,
author = {{Bansal}, N. and {Chen}, X. and {Wang}, Z.},
title = "{Can We Gain More from Orthogonality Regularizations in Training Deep CNNs?}",
journal = {ArXiv e-prints},
archivePrefix = "arXiv",
eprint = {1810.09102},
keywords = {Computer Science - Machine Learning, Computer Science - Computer Vision and Pattern Recognition, Statistics - Machine Learning},
year = 2018,
month = oct,
adsurl = {http://adsabs.harvard.edu/abs/2018arXiv181009102B},
adsnote = {Provided by the SAO/NASA Astrophysics Data System}
}