guarin
guarin
Thanks for the info! We'll check internally if this is an option for us :)
The output dimension of the DINOLoss should correspond to the output dimension of the projection head and not of the backbone. By default it is 65536. For small datasets you...
Depends what you want to do. `pretrain=True` will load very good pretrained weights, this can help you create good embeddings very quickly.
Hi, could you share the relevant code? It looks like the memory bank features need to be transposed. The shapes should be `256x128` and `128x76800`. You can transpose them by...
Hi and thanks for creating the issue! Is there a particular reason why you would prefer it to be centered? As far as I know the default for the Read...
I completely agree that the left alignment doesn't look good on large screens. If there is an easy config option to center it I would consider switching but if it...
Thanks for your interest @Abhrajitdas02 but it is not yet clear if we want to implement this.
Thanks for raising this issue! There is a previous discussion on supporting tensors as transform inputs #791 I think we can switch to torchvision solarization and blur implementations. When making...
Hi yes ofc :) That being said, there are more recent SSL methods that could also be interesting to work on, see https://github.com/lightly-ai/lightly/issues/1172 Maybe LDReg could be a good first...
Hi! It looks like the code in your issue description is not correctly formatted, could you put it into a single markdown codeblock? Regarding the training issues, when exactly does...