[FEATURE] Add DINO v3 weights

[seefun]

DINOv3: Self-supervised learning for vision at unprecedented scale weights for ViT and ConvNeXt released

they provide two categories of SSL dataset: web dataset & satellite dataset

[tms2003]

The attention is almost perfect, which is the most important thing. Strongly request support as soon as possible.

[imneonizer]

Waiting for the model weights.