search

jiepengwang / NeuRIS

MIT License
229 stars 16 forks source link

About the Replica dataset? #31

Open XIE-ZJU opened 8 months ago

XIE-ZJU commented 8 months ago

Hi, thanks for your great job of this work. I want to test your method on replica dataset. I find that you provide the replica_room0 in the one drive. Have you ever preprocessed the other rooms in the replica dataset, can you share the results with me? Thanks again~

jiepengwang commented 8 months ago

Thanks for your interest in our work. Currently, we only preprocessed room0. For other scenes, we suggest to download the pre-rendered data from here (https://github.com/spla-tam/SplaTAM/blob/main/bash_scripts/download_replica.sh) and follow our data preparation guidelines to preprocess the data.

XIE-ZJU commented 8 months ago

Copy that!

XIE-ZJU commented 8 months ago

Oh, thanks for your reply! I just follow your instructions to process the data. I met a problem that the predicted normal result is .png format, how to transfer to .npz format as your provided replica_room0.

XIE-ZJU commented 8 months ago

Apologize for bothering you again! When I trained replica_room2. the code will always break down at renderer.py. File "/root/autodl-tmp/NeuRIS-main/models/renderer.py", line 257, in render_core assert (torch.isnan(variance).any() == False) variance = (1.0 /inv_variance).mean(dim=-1, keepdim=True) assert (torch.isinf(variance).any() == False) assert (torch.isnan(variance).any() == False)

jiepengwang commented 8 months ago

Save the numpy array here to get the .npz file. (https://github.com/baegwangbin/surface_normal_uncertainty/blob/fe2b9f1e8a4cac1c73475b023f6454bd23827a48/test.py#L35). For the exception, I'm not sure whether you preprocess the data correctly.