kiui

Yes, it is a common problem in all dreamfusion-based methods.

@seawater668 Hi, you should merge the audio via ffmpeg manually. Check the `scripts/` or colab for examples.

@ForceConstant It's a good idea, I will use release when there are new improvements.

@JunzheJosephZhu This makes sense! Do you observe the results to be worse compared to the correctly scaled implementation? I guess this may fix it, but need to do some experiments...

Thanks for the information! I have updated the code, maybe you are interested in trying it on your experiments too.

@nazarPuriy Hi, this is strange, maybe you could try using full precision mode? (commenting the `opt.fp16 = True` line in main.py)

@neverix Yes, this is interesting, I guess the pose model can be used to synthesize better human by initializing a skeleton and rendering it from different views as a prior....

@JunzheJosephZhu Maybe you could increase the NeRF rendering resolution (current is 64x64), but it leads to large GPU memory consumption. Another idea is to follow Magic3D, and train a mesh...

@zakkeasterbrook Sorry, this work relies on CUDA environment.

This is strange, what's the environment (e.g., GPU, CUDA, OS) you use? How do you measure the time?