guarin

As @MalteEbner suggested it is often best to not completely freeze the backbone for finetuning on object detection. Regarding the different SSL models you can, for example, find the following...

Thanks @lukasugar for making the issue and investigations! I would suggest the following changes: * Change gaussian blur to torchvision * Add deprecation warning when using our gaussian blur transform...

That sounds great! Thanks a lot, looking forward to your pr :)

The cli still uses SelfSupervisedEmbedding objects. We should decide whether we want to continue using it or also change to low level.

Great find! I completely missed this part of the paper. It looks relatively simple to implement but is not part of the official code which makes it a bit hard...

Yes, I think we can just combine the `MAEBackbone`, a linear prediction head and the l1 loss in a module to build the `SimMIM` model. I would propose to add...

Hi, lightly should be compatible with any recent cuda version as it is built on pytorch. The instructions on how to install pytorch for windows are provided on the pytorch...

> On which dataset are the benchmarks? And which GPU? The GPU memory usage seems waaay off. If you want to add the benchmark results, please make sure you use...

I'll run the 800 epoch benchmark, update the benchmark docs and then merge it. Should be done by tomorrow.

Hi @shree-lily, we sadly did not add this to the documentation yet. You can still use either pytorch or pytorch lightning to save and load checkpoints: * https://pytorch.org/tutorials/beginner/saving_loading_models.html#saving-loading-model-for-inference * https://pytorch-lightning.readthedocs.io/en/stable/common/checkpointing_basic.html...