running on custom datasets
JanaldoChen opened this issue · comments
JanaldoChen commented
Hi, thanks for sharing this wonderful work! Can you release the code for custom datasets like the aist++ dataset mentioned in your paper?
Youngjoong Kwon commented
Hello!
I am currently very busy with other projects, so could you refer to the Neural Body github regarding the custom dataset for the moment? Our dataset structure follows the Neural Body dataset structure.
In the current version of our code, it requires rasterization results, but you can modify the code to not use them as follows (I will also update the code to support the version that does not require rasterization results):
- Set the 'rasterize' and 'use_viz_test' as 'False' in the configuration file located at 'configs' folder
- In the renderer file, delete L68-69 and modify L71 as 'return torch.ones_like(latent.shape[:2]), latent'