I might be setting the bar too high, but the exact problem I'm trying to solve is reconstructing a moving object on a conveyor belt using images from fixed cameras. In my test setting, I have 30 good images per camera, with three cameras situated around the conveyor belt. So not that many images.
Got it in my scholar inbox this morning.
arxiv.org/abs/2502.12138
Great work! I'm investigating if the recent work in 3D reconstruction (Mast3r, 3DGS...) could be used for Industrial anomaly detection. As in precise enough metrology on the reconstructed object. Do you think your work would be a good fit?