search

DepthAnything / Depth-Anything-V2

Depth Anything V2. A More Capable Foundation Model for Monocular Depth Estimation
https://depth-anything-v2.github.io
Apache License 2.0
2.56k stars 186 forks source link

Teacher model #59

Open toanbh10032 opened 2 weeks ago

toanbh10032 commented 2 weeks ago

Can you release the "trained teacher model" which used to create pseudo GT for real image?

LiheYoung commented 1 week ago

The teacher model is based on DINOv2-Giant. Unfortunately, we are still undergoing company review process to release the most capable model.

toanbh10032 commented 1 week ago

The teacher model is based on DINOv2-Giant. Unfortunately, we are still undergoing company review process to release the most capable model.

I mean the Dinov2 model, not the vitl model. Because I want to use it to make dataset to re-train the model, or if you can public the dataset and groundtrtuth that you used to train student model?

LiheYoung commented 2 days ago

We used public datasets (e.g., SA-1B). You can use our released Depth-Anything-V2-Large model to assign pseudo labels to these raw images to train your models.