Applying Label Wise Attention for 2 datasets with extreme label space (>4k labels) and imbalance (including zero-shot): Eurlex-57K and StackExchange.
Comparing and optimizing CNN (ResNet-50) and ViT (ViT-B/16) architectures on the FGVC-Aircraft dataset.
Optimizing CLIP for food classification on the Food-101 dataset, comparing zero-shot and few-shot performance, and analyzing failure cases.