One Block, Multiple Depths: Recurrent Vision Transformers with Depth-Programmed Experts

作者：Adrian Bulat, Yassine Ouali, Georgios Tzimiropoulos · 会议/期刊：arXiv preprint · 方向：cs.CV · 发布日期：2026-10-09