How to continue self-supervised pre-training on my own dataset?

[RalphCBY]

Thanks for this work!

I have some questions regarding continuing pre-training:

1. My dataset consists of satellite images I noticed you used 512x512 images and set global_size to 256x256. I'd like to know what local_size you configured and whether you could share the config file for satellite image pre-training to provide more details about the training strategy?

2. Additionally, how should I load the ‘dinov3-vitl16-pretrain-sat493m’ weights when continuing pretraining? Is it loaded in the config file? I only see load_checkpoint set in the student section. Or should I replace the weight initialization in train.py with reading the weight file?

Thank you for your help.