The model reaches a maximum accuracy of 91.51% in 47 epochs on CIFAR-10 dataset using ResNet-18 model.
- Loss Function: Cross Entropy Loss (combination of
nn.LogSoftmaxandnn.NLLLoss) - Optimizer: SGD
- Learning Rate: 0.01
- LR Step Size: 25
- LR Gamma: 0.1
- Batch Size: 64
- Epochs: 50
The following data augmentation techniques were applied to the dataset during training:
- Random Horizontal Flip: 0.2
- Random Vertical Flip: 0.1
- Random Rotation: 10 degrees
- Random Erasing: 0.3
The transforms library in the package torchvision was used to apply augmentation.
Install the required packages
$ pip install -r requirements.txt
Upload the files in the root folder and select Python 3 as the runtime type and GPU as the harware accelerator.
- Shantanu Acharya (Canvas ID: 25180630)
- Rakhee (Canvas ID: 25180625)