Cut-Thumbnail (Accepted at ACM MULTIMEDIA 2021)

Tianshu Xie, Xuan Cheng, Xiaomin Wang, Minghui Liu, Jiali Deng, Tao Zhou, Ming Liu

This is the official Pytorch implementation of Cut-Thumbnail in the paper Cut-Thumbnail:A Novel Data Augmentation for Convolutional Neural Network.

This implementation is based on these repositories:

Main Requirements

  • torch == 1.0.1
  • torchvision == 0.2.0
  • Python 3

Training Examples

  • Mixed Single Thumbnail

python train.py -d [datasetlocation] --depth 50 --mode mst --size 112 --lam 0.25 --participation_rate 0.8
  • Self Thumbnail

python train.py -d [datasetlocation] --depth 50 --mode st --size 112 --lam 0.25 --participation_rate 0.8

Results

  • ImageNet Results
Model Accuracy (%)
ResNet50 + CutMix *
ResNet50 + Cut-Thumbnail (ST) 77.74
ResNet50 + Cut-Thumbnail (MST) 79.21

* denotes results reported in the original papers.

  • CIFAR-100 Results
Model Accuracy (%)
WideResNet-28-10 + Cut-Thumbnail (ST) 81.41
WideResNet-28-10 + Cut-Thumbnail (MST) 83.35
  • CUB-200-2011 Results
Model Accuracy (%)
ResNet50 + Cut-Thumbnail (ST) 85.72
ResNet50 + Cut-Thumbnail (MST) 86.56
ResNet50 + Cut-Thumbnail (MDT) 86.72

Citation

If you find our paper and this repo useful, please cite as

@inproceedings{xie20cut-thumbnail,
    author = {Xie, Tianshu and Cheng, Xuan and Wang, Xiaomin and Liu, Minghui and Deng, Jiali and Zhou, Tao and Liu, Ming},
    title = {Cut-Thumbnail: A Novel Data Augmentation for Convolutional Neural Network},
    year = {2021},
    isbn = {9781450386517},
    publisher = {Association for Computing Machinery},
    address = {New York, NY, USA},
    url = {https://doi.org/10.1145/3474085.3475302},
    doi = {10.1145/3474085.3475302},
    booktitle = {Proceedings of the 29th ACM International Conference on Multimedia},
    pages = {1627–1635},
    numpages = {9},
    location = {Virtual Event, China},
    series = {MM '21}
}

GitHub

View Github