OzzyGT/MoGe_2_ViTL

MoGe 2 ViT-L

Safetensors conversion of MoGe2 ViT-L for monocular geometry and camera field-of-view estimation. Weights are stored in FP16 (about 652 MB), with the original tensor names and shapes preserved. This is the 326M-parameter model without the optional surface-normal head.

Source and license

Converted from model.pt at revision 39c4d5e957afe587e04eec59dc2bcc3be5ecd968. Weights were cast from FP32 to FP16; no training was performed. Licensed under MIT. See LICENSE.