moge-3-vitg-torch#

MoGe-3 ViT-g (1.25B): the largest MoGe-3 model for metric monocular geometry, with depth, a camera-space point map, surface normals and camera intrinsics, refined by a sparse volumetric stage.

Details

  • Model name: moge-3-vitg-torch

  • Model source: microsoft/MoGe

  • Model author: Lingyu Kong, Ruicheng Li, Ruicheng Wang, et al.

  • Model license: MIT

  • Model size: 4.66 GB

  • Exposes embeddings? no

  • Tags: depth, 3d, torch, transformer

Requirements

  • Packages: torch, torchvision

  • CPU support

    • yes

  • GPU support

    • yes

Example usage

 1import fiftyone as fo
 2import fiftyone.zoo as foz
 3
 4dataset = foz.load_zoo_dataset(
 5    "coco-2017",
 6    split="validation",
 7    dataset_name=fo.get_default_dataset_name(),
 8    max_samples=50,
 9    shuffle=True,
10)
11
12model = foz.load_zoo_model("moge-3-vitg-torch")
13
14dataset.apply_model(model, label_field="predictions")
15
16session = fo.launch_app(dataset)