Thanks to visit codestin.com
Credit goes to github.com

Skip to content

Commit b942f00

Browse files
EasternJournalistEasternJournalist
authored
Update V2 training (#118)
### Added - Update training code for MoGe-2. ### Changed - Refactored training dataloader code for better readability. - Removed Git LFS for convenience.
1 parent 0286b49 commit b942f00

39 files changed

+861
-1048
lines changed

‎.gitattributes‎

Lines changed: 0 additions & 2 deletions
This file was deleted.

‎CHANGELOG.md‎

Lines changed: 9 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -29,4 +29,12 @@
2929

3030
## 2025-06-10
3131
### Added
32-
- Released MoGe-2.
32+
- Released MoGe-2.
33+
34+
## 2025-10-16
35+
### Added
36+
- Update training code for MoGe-2.
37+
38+
### Changed
39+
- Refactored training dataloader code for better readability.
40+
- Removed Git LFS for convenience.

‎README.md‎

Lines changed: 9 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -35,6 +35,9 @@ https://github.com/user-attachments/assets/8f9ae680-659d-4f7f-82e2-b9ed9d6b988a
3535

3636
## ✨ News
3737

38+
***(2025-10-16)***
39+
* Updated training code for MoGe-2.
40+
3841
***(2025-06-10)***
3942

4043
* ❗**Released MoGe-2**, a state-of-the-art model for monocular geometry, with these new capabilities in one unified model:
@@ -272,14 +275,12 @@ See [LICENSE](LICENSE) for more details.
272275
If you find our work useful in your research, we gratefully request that you consider citing our paper:
273276

274277
```
275-
@misc{wang2024moge,
276-
title={MoGe: Unlocking Accurate Monocular Geometry Estimation for Open-Domain Images with Optimal Training Supervision},
277-
author={Wang, Ruicheng and Xu, Sicheng and Dai, Cassie and Xiang, Jianfeng and Deng, Yu and Tong, Xin and Yang, Jiaolong},
278-
year={2024},
279-
eprint={2410.19115},
280-
archivePrefix={arXiv},
281-
primaryClass={cs.CV},
282-
url={https://arxiv.org/abs/2410.19115},
278+
@inproceedings{wang2025moge,
279+
title={Moge: Unlocking accurate monocular geometry estimation for open-domain images with optimal training supervision},
280+
author={Wang, Ruicheng and Xu, Sicheng and Dai, Cassie and Xiang, Jianfeng and Deng, Yu and Tong, Xin and Yang, Jiaolong},
281+
booktitle={Proceedings of the Computer Vision and Pattern Recognition Conference},
282+
pages={5261--5271},
283+
year={2025}
283284
}
284285
285286
@misc{wang2025moge2,

‎assets/normal_comaprison.jpg‎

2.43 MB
Loading

‎assets/overview_simplified.png‎

404 KB
Loading

‎assets/panorama_pipeline.png‎

721 KB
Loading

‎baselines/moge.py‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -40,7 +40,7 @@ def load(num_tokens: int, resolution_level: int, pretrained_model_name_or_path:
4040
@torch.inference_mode()
4141
def infer(self, image: torch.FloatTensor, intrinsics: Optional[torch.FloatTensor] = None):
4242
if intrinsics is not None:
43-
fov_x, _ = utils3d.torch.intrinsics_to_fov(intrinsics)
43+
fov_x, _ = utils3d.pt.intrinsics_to_fov(intrinsics)
4444
fov_x = torch.rad2deg(fov_x)
4545
else:
4646
fov_x = None
@@ -62,7 +62,7 @@ def infer(self, image: torch.FloatTensor, intrinsics: Optional[torch.FloatTensor
6262
@torch.inference_mode()
6363
def infer_for_evaluation(self, image: torch.FloatTensor, intrinsics: torch.FloatTensor = None):
6464
if intrinsics is not None:
65-
fov_x, _ = utils3d.torch.intrinsics_to_fov(intrinsics)
65+
fov_x, _ = utils3d.pt.intrinsics_to_fov(intrinsics)
6666
fov_x = torch.rad2deg(fov_x)
6767
else:
6868
fov_x = None

0 commit comments

Comments
 (0)