Can we directly finetune other multi-view imgaes version based on existing ckpt?

[XiaoKangW]

Dear Sir，

Thanks for your great job and open source your code.

And i want to ask whether we can directly finetune other multi-view imgaes version based on existing ckpt? For example, we finetune 3 views on 2 views version. Or we only trainning 3-view from scratch?

If we train with 3 views, how many scenes data entries would be sufficient, what size of GPU is needed, and how long does it usually take to complete the training?

We look forward to your reply, as it is crucial for ours. Thank you!

Best

[Xiangyu1Sun]

We tried to train the model from 2-6 views input and get meaningful and great performance when do inference. To train 3 views, it would need 8 L20-48GB gpus and the training of the model will finish in 3 days.

[XiaokangWei]

Thanks for your reply!

And I want to ask whether you can release your ckpt in 3 views in your future plan.