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

how to use input camera extrinsic correctly

Open
#192 2 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, da-3 team,

I am doing some experiments with da-3 on [ml-hypersim](https://github.com/apple/ml-hypersim), as Iexpected with input camera intrinsics and extrinsic, the recon quality should be more better. but somehow looks after inject camera extrinsic, it got very poor alignment issue:

baseline: scene ai_001_001 rgbs + cam intrins only

Image

once added cam extrinsic on the baseline:

Image

btw, I am using the [script](https://github.com/ZJLi2013/ml-hypersim/blob/main/scripts/hypersim_camera_depth.py) to extract camera extrinc from ml-hypersim dataset

and a simple [sample](https://github.com/ZJLi2013/Depth-Anything-3/blob/main/scripts/basic_gs.py) used to test

test as

```sh
python3 basic_gs.py --input-images /workspace/dataset/hypersim/color/input_images --intrinsics-json /workspace/dataset/hypersim/ai_001_001/_detail/cam_00/intrinsics.jsonl --extrinsics-jsonl /workspace/dataset/hypersim/ai_001_001/_detail/cam_00/extrinsics.jsonl --output-dir /workspace/dataset/hypersim/ai_001_001/output
```

any suggestions why the extrinsic got messed ?

Thanks a lot
ZJ

Contributor guide

No contributing guide indexed for this repository

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.