The model reaches a test accuracy of 84.83% in CIFAR-10 dataset. The model uses the following types of convolutions:
- 3x3 Convolution
- Pointwise Convolution
- Atrous Convolution
- Depthwise Separable Convolution
- Max Pooling
The model has 94,218 parameters.
- Loss Function: Cross Entropy Loss
- Optimizer: SGD
- Learning Rate: 0.01
- Dropout Rate: 0.1
- Batch Size: 64
- Epochs: 50
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.
- Rakhee (Canvas ID: 25180625)
- Shantanu Acharya (Canvas ID: 25180630)