ICCV-2021 Emerging Properties in Self-Supervised Vision Transformers 2022-02-04 PaperNote CV, cs.Transformer 0 0 Comments Word Count: 89(words) Read Count: 1(minutes) 论文地址:Emerging Properties in Self-Supervised Vision Transformers DINO:融入ViT的对比学习,不用标签训练的自注意效果好 不用标签训练的自注意图效果极好,媲美直接对物体做语义分割 防止训练坍塌,对teacher网络进行centering,指把一个batch作均值后减掉,防止模型坍塌