VISION TRANSFORMERS FOR DENSE PREDICTION TASKS

Summary from the TLDR AI newsletter

This project introduces ViT-CoMer, a new backbone architecture enhancing Vision Transformers (ViT) for dense prediction tasks without the need for pre-training.