sabin5105/CIFAR10

Various model comparison for CIFAR10

CIFAR10

various models adaption for CIFAR10 dataset

Models

• Decision Tree
• Random Forest
• Logistic Regression
• SVM (One-vs-Rest)
• CNN (basic custom)
• ResNet18 (scratch)
  • model/Resnet.py
• ResNet50 (pretrained)
  • Backbone : ImagenetV2
  • torchvision.models
• ESRGAN
  • srgan/esrgan/models.py
• EfficientNet
• add more soon

Dataset

• Information : https://www.cs.toronto.edu/~kriz/cifar.html
• shape:
  • training : (50000, 32, 32, 3)
  • test : (10000, 32, 32, 3)
• class : 10
• example

• distribution

Data augmentation

• RandomCrop
• RandomHorizontalFlip
• RandomVerticalFlip
• RandomRotation
• super resolution
• interpolation

• add more soon

Result

• table will be here

reference papers

• Ashish Vaswani, Attention is all you need, 2017(transformer)

  • https://arxiv.org/abs/1706.03762

• Lia deng, ImageNet: A Large-Scale Hierarchical Image Database, 2009

  • https://ieeexplore.ieee.org/document/5206848

• kaiming he, 2016, Deep Residual Learning for Image Recognition (ResNet)

  • https://arxiv.org/abs/1512.03385

• ryo takahashi, 2019, Data Augmentation using Random Image Cropping and Patching for Deep CNNs (Random Image Cropping)

  • https://arxiv.org/abs/1906.11172

• Connor Shorten, 2019, A survey on Image Data Augmentation for Deep Learning (Data Augmentation - flip, rotate, crop, etc)

  • https://journalofbigdata.springeropen.com/articles/10.1186/s40537-019-0197-0

• Christian Ledig, 2016, Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network (SRGAN)

  • https://arxiv.org/abs/1609.04802

• Xintao Wang, 2018, ESRGAN: Enhanced Super-Resolution Generative Adversarial Networks (ESRGAN)

  • https://arxiv.org/abs/1809.00219