lakshmi narayan comments

Repositories
Issues
Comments

Results 3 comments of


                                            lakshmi narayan

What hyperparameters should be changed for effective training on Meta-Dataset subset?

Hi, Thanks for quick reply. My backbone SSL model is vit-tiny (~5.7 million parameter model). The following are the meta-training args that I used. num_epochs: 100 image_size: 128 Num ways:...

about rotate

Can someone please explain what is the right format for the rotation angles. Also, why are absolute values taken for the angles (torch.abs(angles)). If I am trying to synthesize right...

Model weights license

Thanks for the quick reply. The [paper](https://arxiv.org/pdf/2108.04539) states that for pre-training, the IIT-CDIP dataset (excluding 400k RVL-CDIP ) was used. I believe this dataset is public.