About this document

Enhancing ResNet50 with U-Net Attention by Harsha Bomanna is a document available to read on EtoBox.

This task involves fine-grained image classification of product images. The baseline model is ResNet50 without pretraining. Two proposed models add a U-Net based attention branch on top of ResNet50. Model 1 uses a single U-Net+++ to generate attention maps for all feature maps. Model 2 uses separate mini U-Net+++ for each feature map, connected by skip connections. Both proposed models achieved better accuracy than the baseline on the test set, with Model 1 performing best with an accuracy of 91.73%.

Author

Harsha Bomanna

Language

EN