Closed z050209 closed 9 months ago
Hello @z050209 ,
I believe our method can be adjusted to suit various GPUs. You have the flexibility to fine-tune several hyperparameters for this adaptation. First, the number of points used to represent the geometry of each object. Second, the batch size employed for model training. If you encounter VRAM constraints, consider reducing these parameters, and you can identify the optimal parameters for running our pipeline.
Best
Thanks, I will try that. 👍
Dear authors, may I know which GPU environment would be sufficient to train this model, if using GRAB or ARCTIC dataset?
thanks.