[question] More close up images lead to worse result

[tak-ho]

Describe the problem I originally use 724 drone images to generate the building 3d model and it looks ok Although still many holes. But then if I add 2787 more close up images and generate all over it become worse, the 3d model broken then before.

Screenshots 724 images result:

724+2787 images result:

Desktop (please complete the following and other pertinent information):

• OS: Ubuntu
• Python version: 3.9.12
• Meshroom version: Meshroom-2023.2.0-av3.1.0-centos7-cuda11.3.1 (binary/prebuild)

[tak-ho]

I noticed people mention about Augment Reconstruction, should I try this approach? But I still don't know why adding images will lead to worse result, I through non-calibrated camera/image will just get ignored?

[natowi]

The repeating patters of the building surface can be a problem while matching the close ups to the larger scene. You could try if using your gps data can help: https://github.com/alicevision/Meshroom/wiki/Model-orientation also maybe something like https://github.com/alicevision/Meshroom/issues/2132#issuecomment-1666898231

Finding the best workflow for you may take some time and experimentation.

[fabiencastan]

Could you share your dataset privately to reproduce your issue?

[tak-ho]

Could you share your dataset privately to reproduce your issue?

Could you please provide way to send you private message? I cannot do it in github and cannot find your correspondent email in github too, thanks

[natowi]

You can contact the core team privately on: [email protected].