search

hustvl / 4DGaussians

[CVPR 2024] 4D Gaussian Splatting for Real-Time Dynamic Scene Rendering
https://guanjunwu.github.io/4dgs/
Apache License 2.0
2.25k stars 186 forks source link

Render From specific viewpoint #137

Open m4ri01 opened 6 months ago

m4ri01 commented 6 months ago

Hello, and thank you for your excellent work on these 4DGaussians. I really appreciate this method and code. I want to ask about the rendering process for the Dynerf dataset. In the dynerf dataset (i.e., cut_roasted_beef), if I render it using this commandpython render.py --model_path "output/dynerf/cut_roasted_beef/" --skip_train --configs arguments/dynerf/cut_roasted_beef.py  & I will get a result that looks like yours on the project web: https://guanjunwu.github.io/4dgs/static/videos/cut_roasted_beef_2_30.mp4. My question is: how do I render the train result so that I can only view it from a specific angle? For example, I want to render the result from the left perspective. I expect that the output should closely resemble the cam10.mp4 found in the original cut_roasted_beef data set. Is it possible to do it this way?

guanjunwu commented 5 months ago

Hi，you can select a training view and rotate it? just modify the code in here

Spawnfile commented 4 months ago

According to https://github.com/hustvl/4DGaussians/issues/137#issuecomment-2187862008, does view like get_spiral affect to train result or only render output ?