Professor Announces Student Presentation on AFFMAE

Fuxin Li, an associate professor at Oregon State University, stated in a post that their AutoFocusFormer from CVPR 2023 learns to downsample to the most important tokens in an image. He noted that implementation efficiency has limited gains mostly to FLOPS benefits rather than actual speed. Li added that his student David Smerkous will present AFFMAE at ECCV 2026. The post included a 125-second video of presentation slides that opens with the title AFFMAE: AutoFocusFormer Masked AutoEncoder.