Finetuning for a specific scenarios

[wowkie83]

thank you for the great work!!! :)
my question is:
If I only want to improve the performance of the model in a very fixed scenario using the same camera, the same distance, and very similar objects, is it enough to use a dataset of 1000 pairs for fine-tuning?
And do I need to convert my own dataset into Habitat format because there is a bug in the co3d code? I want to confirm whether habitat code can generate good models?

thnaks