ByteDance-Seed / ByteDance-Seed/Depth-Anything-3

Recover the road 3D model with DA3, but the metric depth looks wrong

Open
#94 5 comments 0 reactions 0 assignees View on GitHub
Dominant language
Python
Stars
6.3k
Forks
702
PR merge metrics
No merged PRs in 30d

Description

Hi,DA3 is a great job. I want to see if I can recover the real scale 3D model of the road through DA3 by inputting the vehicle dashcam image. The depth information of each image looks good, but there seems to be a big bias in using the intrinsics to recover the true scale. I think it's a lot bigger than it really is.
The following is my code, can you give me some advice, thank you very much!

Image

Image

Image

Image

Contributor guide

No contributing guide indexed for this repository

Research direction

Start by reviewing the code shown in the attached images and the DA3 metric-depth and camera-intrinsics guidance. Reproduce the road-scene reconstruction and compare the supplied intrinsics with the recovered scale; done means identifying why the result is oversized or documenting the calibration required.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
computer-vision, machine-learning
Issue type
Bug
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
15/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.