search

hustvl / 4DGaussians

[CVPR 2024] 4D Gaussian Splatting for Real-Time Dynamic Scene Rendering
https://guanjunwu.github.io/4dgs/
Apache License 2.0
2.23k stars 182 forks source link

Why does the decoding of the dynerf dataset start from cam01.pm4, not cam00.pm4？ #99

Closed yanh929 closed 7 months ago

yanh929 commented 8 months ago

I am a beginner and I apologize for starting this issue. I hope you can provide assistance.

yanh929 commented 8 months ago

using "python train.py -s data/dynerf/sear_steak --port 6017 --expname "dynerf/sear_steak" --configs arguments/dynerf/sear_steak.py"

guanjunwu commented 8 months ago

Hi, cam00.mp4 is used for evaluation, 01-20 are used for training.

yanh929 commented 8 months ago

@guanjunwu After decoding the frame, I obtained images from 01-20 and removed MP4. Executing "bash colmap. sh data/dynerf/sea_steak llff", there is a discrepancy between the number of files and the number of cameras in npy. Do we need to modify the npy file?

jsxzs commented 8 months ago

@yanh929 cam00.mp4 should also be decoded. You can look at the new PR, which provides instructions for training dynerf scenes.

yangqing-yq commented 7 months ago

@yanh929 cam00.mp4 should also be decoded. You can look at the new PR, which provides instructions for training dynerf scenes.

for customized dataset, how to generate poses_bounds.npy file? @guanjunwu @jsxzs