SelfSup-MIM

Hi, we haven't met this problem before and I guess it has nothing to do with the code. Are the environment installed exactly the same as the readme file?

Sorry that we did not save the ViT-S checkpoint😢

Hi, it is ok to use just the encoder. The so-called `teacher model` is actually the same as the encoder (structure/parameter).

It is ok to do so. This codebase defines a `teacher model` class to provide other researchers with the possibility to use EMA, although EMA is not used in CAE.

Hi, we did not save the detection model. I think the code could reproduce the results well. If there is some question, please tell me~

@TinnyFlames @Yindong-Zhang @w5yang Hi, please leave your email address here. I can send you a full checkpoint (ViT-base, 1600 epochs).

> @SelfSup-MIM ViT-base, 1600 epochs 的完整模型参数可以也发我一下吗，非常感谢。邮箱[[email protected]](mailto:[email protected]) 已发~

> @SelfSup-MIM 你好也可以给我发一下ViT-base 1600epochs的完整模型参数吗，非常感谢，邮箱[[email protected]](mailto:[email protected]) 已发~

@Vickeyhw @echoanran I have sent the full checkpoint of ViT-base to your emails. I don't have full checkpoint of ViT-large at hand.😢

Hi, there are several tips that may help alleviate the issue. 1. Decrease learning rate. 2. Enlarge the drop path ratio. 3. Decrease clip grad number. 4. Enlarge warm up...