Cannot reconstruct on very long video sequence.

[LaFeuilleMorte]

Hi, I've tried to use spann3r on a very long video (about 3 minutes). The video was an indoor scene that scanned all the rooms in that house. But the reconstructed results seemed to have only reconstructed one of the rooms in the whole house. The other rooms are missing. Can I finetune on my datasets to improve the performance on such scenes. Or I should use other techniques like (visual slam or vins slam)?

[HengyiWang]

Hi @LaFeuilleMorte, as discussed in 4.4, our model may not be able to reconstruct a multi-room scene due to the limited spatial region that training frames can cover. You can try to restart the spann3r every few frames and align those fragments together.