Single view?

[AlexMorgand]

Hey it's me again :)

I tried to use the method for only one view as shown in the paper without any success :( (mast3r-sfm probably needs a pair here and even duplicating the first frame didn't work

Anything specific that I need to do? Thanks again!

[DaLi-Jack]

Hi! I really appreciate you taking the time to test the method so thoroughly. I’m always happy to help and discuss anything, so feel free to reach out anytime.

Regarding your question: for single-view experiments, we rely on camera-controlled diffusion models such as ZeroNVS and Stable Virtual Camera. These models first generate multi-view images from the single input view. Then MASt3R-SfM is run on those generated multi-view images, not directly on the single original frame. The full details are described in Appendix A.1 and Appendix C.4 of our paper. Additionally, you can also use image-to-video models like Google Veo 3.1 and Kling AI to generate multi-view sequences from a single image, since our method also supports unposed image inputs.

Hope this helps! And again, feel free to ask anytime — happy to support you however I can.