DUSt3R-to-COLMAP

Pulling data from DUSt3R to fit the COLMAP output format.

Preparation

wget -nc https://download.europe.naverlabs.com/ComputerVision/DUSt3R/DUSt3R_ViTLarge_BaseDecoder_512_dpt.pth -P checkpoints/

git clone https://github.com/naver/dust3r --recursive

pip install -r requirements.txt

Make dir 'scene_name' in data/images/ Download image to scene_name folder

Run Code

Directory setting is needed before running the code

dust3r_inference.py

Data directory

• Source images are saved under /data/images with each folder named after the scene_name.
• For each scene_name, a directory /data/scenes/scene_name is created, and inside it, the required COLMAP output data is generated in the sparse/0 folder.