项目作者: sousquared

项目描述 :
CNN training strategy with blurred images
高级语言: Jupyter Notebook
项目地址: git://github.com/sousquared/blur-training.git
创建时间: 2021-02-04T13:29:50Z
项目社区:https://github.com/sousquared/blur-training

开源协议:MIT License

下载


blur-training

Training CNNs(Convolutional Neural Networks) with blurred images.

Sou Yoshihara, Taiki Fukiage, Shin’ya Nishida, “Does training with blurred images bring convolutional neural networks closer to humans with respect to robust object recognition and internal representations?”, Front. Psychol., Vol. 14,2023

Blurred images (GaussianBlur)

Training images are blurred by Gaussian function. The images are more blurred as std.(σ) of Gaussian kernel is bigger.
blurred-images

Schedule

You can try different training schedule as for blurring images. Here is an overview of the training schedule:

(In ./cifar10/, the terms are different, and some modes may not be included.)
schedule

Training

Datasets

(Note that you have to download ImageNet by yourself.)

  • ImageNet dataset
  • 16-class-ImageNet dataset
    This dataset is from Geirhos et al., 2018.

    (R. Geirhos, C. R. M. Temme, J. Rauber, H. H. Schütt, M. Bethge and F. A. Wichmann: Generalisation in humans and deep neural networks. Advances in Neural Information Processing Systems (NeurIPS), 7538–7550, 2018.)

    I make the dataset from ImageNet by using robustness library.
  • Cifar-10 dataset

References

pytorch tutorial

pytorch imagenet training example