ImageClassification-13KenyanFood-PyTorch

Image Classification Task which is part of the Deep Learning with PyTorch OpenCV course.

The dataset consists of 8,174 images in 13 Kenyan food type classes from which the 20% was used for validation.

The input image size is 224 by 224 pixels and I have used several Data Augmentation techniques, such as Random Vertical/Horizontal Rotation and Color Jitter (brightness, contrast).

I have used pretrained Resnet50 (wide_resnet50_2) and froze 60% of the parameters, leaving the last 40% of the network parameters trainable. The model was trained for 30 epochs and achieved 94% training and 75% validation accuracy. The highest score of my Kaggle submission is currently at 73.5% accuracy.

Initial training Tensorboard logs: https://tensorboard.dev/experiment/VtMb0okWSXCqv3nPSZyHRw/

-Todo

1. Modularize the training pipeline to allow for rapid experimentation.
2. Refactor code to increase readability.
3. Try ensemble learning to achieve higher accuracy.
4. Update tensorboard logs.