Skip to content
master
Switch branches/tags
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
 
 
 
 
 
 
 
 
 
 
 
 

ADA-Net

Tensorflow implementation

Semi-Supervised Learning by Augmented Distribution Alignment Qin Wang, Wen Li, Luc Van Gool (ICCV 2019 Oral)

Thesis: Distribution Aligned Semi-Supervised Learning 2018 August at ETH Zurich

Requirements

pip3 install tensorflow-gpu==1.13.1
pip3 install tensorpack==0.9.1
pip3 install scipy==1.2.1

Train and Eval ADA-Net on ConvLarge

Prepare dataset

cd convlarge
python3 cifar10.py --data_dir=./dataset/cifar10/ --dataset_seed=2

Train and Eval ADA-Net on Cifar10 ConvLarge

CUDA_VISIBLE_DEVICES=0 python3 train_cifar.py --dataset=cifar10 --data_dir=./dataset/cifar10/ --log_dir=./log/cifar10aug/ --num_epochs=2000 --epoch_decay_start=1500 --aug_flip=True --aug_trans=True --dataset_seed=2
CUDA_VISIBLE_DEVICES=0 python3 test_cifar.py --dataset=cifar10 --data_dir=./dataset/cifar10/ --log_dir=<path_to_log_dir> --dataset_seed=2

Here are the error rates we get using the above scripts :

Data Split Seed 1 Seed 2 Seed 3 Reported
8.61% 8.89% 8.65% 8.72+-0.12%

The dataset split seed controls the split between labeled and unlabeled samples. It does not affect the test set.

Train and Eval ADA-Net on ImageNet ResNet

Download our imagenet labeled/unlabeled split from this link, put them in ./resnet

cd resnet
python3 ./adanet-resnet.py --data <path_to_your_imagenet_files> -d 18  --mode resnet --batch 256 --gpu 0,1,2,3

Acknowledgement

  • ConvLarge code is based on Takeru Miyato's tf implementation.
  • ResNet code is based on Tensorpack's supervised imagenet training scripts.

License

MIT

Citing this work

@article{wang2019semi,
  title={Semi-Supervised Learning by Augmented Distribution Alignment},
  author={Wang, Qin and Li, Wen and Van Gool, Luc},
  journal={arXiv preprint arXiv:1905.08171},
  year={2019}
}

Reproduce Figure 4

To reproduce Figure 4 in the paper, we provide the plot script and extracted features here. Notice that we use sklearn==0.20.1 for TSNE calculation.

About

[ICCV 2019 oral] Code for Semi-Supervised Learning by Augmented Distribution Alignment

Resources

License

Packages

No packages published

Languages