much slower than dreamgaussian
EfficientDreamer opened this issue · comments
Thanks for your perfect work! I wonder why this implementation is much slower than Dreamgaussian at https://github.com/dreamgaussian/dreamgaussian/tree/main. Can we reduce the training step to 500 like Dreamgaussian ?
Hi @EfficientDreamer ,
I think the efficiency of DreamGaussian comes from two key components. The first is the Gaussian splatting representation, and the second is the proposed mesh optimization technique. In DreamGaussian, Gaussian splatting is utilized solely for generating the coarse model, enabling 3D generation in just 500 iterations. In the future, I will consider implementing the mesh optimization in DreamGaussian.