Some questions about custom datasets
mojingtong opened this issue · comments
937299 commented
Hello~When I use my own datasets for training, I have several questions:
- I use my own video for preprocessing and training. Then I get the human model and scene model. But when I try to render reposing, the rendering effect is not as good as that in the paper. The rendering pictures are a little bit vague. Is it related to the parameters that I set in trainging? For example, samples_per_ray, rays_per_batch or any other parameter?
- When I try to use other datasets not SFU datasets for render reposing, are there any parameters that I need to adjust in render_reposing.py except for parameters in get_manual_alignment function?
Thanks for your reply!
Wei Jiang commented
Hi~
- You can also check the canonical rendering, if the canonical renderings are also blurry, it means the human model is not well trained. You can load the current weight as a starting point and restart the training, sometimes this retraining will give you a better human reconstruction.
- No, there are no other parameters other than get_manual_alignment, get_manual_alignment is the only params you need to provide.