Learning Object Focused Attention

Vivek Trivedy, Amani Almalki, Longin Jan Latecki

We propose an adaptation to the training of Vision Transformers (ViTs) that allows for an explicit modeling of objects during the attention computation. This is achieved by adding…