About training dataset

[LaFeuilleMorte]

Hi., thanks for your great work. Your idea is very impressive. From my perspective, Reconstruction is a semantic-less work. We only needs image matching, and RGB fit to construct the 3D scene. Like we do colmap and 3DGS optimization. So is it possible to train on massive scale synthetic dataset to further boosting your model?

[haofeixu]

Hi, thank you for your kind words. Yes, I agree with you and our model can be further boosted with more training datasets, longer training etc.