search

ingra14m / Deformable-3D-Gaussians

[CVPR 2024] Official implementation of "Deformable 3D Gaussians for High-Fidelity Monocular Dynamic Scene Reconstruction"
https://ingra14m.github.io/Deformable-Gaussians/
MIT License
885 stars 50 forks source link

fid does not match frame #69

Open cdfan0627 opened 2 months ago

cdfan0627 commented 2 months ago

您好，我在用NeRF-DS dataset training的時候發現，每個frame的uid與fid不匹配，如下圖，可以請問一下原因嗎

image image

在as_novel_view的class理論上uid = 54 ，fid應該要等於54 / 845 = 0.064才對，不應該是上面的0.3290

ingra14m commented 2 months ago

Hi, thanks for your interest.

I found that in NeRF-DS dataset, the uid is not consistent with the actual frame id. So I used the image name as the fid to obtain the right sequence, which is shown in the code here.

cdfan0627 commented 2 months ago

可是NeRF-DS dataset 的 fid不是定義在這裡嗎https://github.com/ingra14m/Deformable-3D-Gaussians/blob/main/scene/dataset_readers.py#L459

ingra14m commented 2 months ago

sry for the wrong link. What I want to say is that the uid in NeRF-DS dataset is not consistent with the actual frame id. That's why in metadata.json, there is a dedicated field defined as time_id.