Need details(?)
#9
by
akzsh
- opened
Hey,
Is there any detail of the model available? Is it pre-trained on WebLI same as SigLIP (Zhai, et, al 2023) or different dataset. Is it a finetune of original SigLIP(?)
Any resources on this would be helpful.
Thanks
It's the same open weights of SigLIP, but with additional positional embeddings (to go to resolution 980) and the processing of images in Navit style (not resized in square). It should be used for a further training