brain-tumor-classifier - A Mixture of experts approach

Using pytorch's TIMM package to solve a classification problem

Details of training

• Training and Testing, separate datasets found at kaggle
• Using 2 Experts, with a backbone supported by SwinTransformer (previously it was efficientnet_b2)
• Batch of 32 images
• Some image transformations
• Advanced training loops

MoE Architecture

Results