Rendering video is blank

[zhuyf66]

inputs:

outputs:

settings:
python train.py --configs arguments/i2v.py -e A_panda_dancing_big_motion --name_override A_panda_dancing_big_motion

python render_for_eval.py --id="A_panda_dancing_big_motion" --savedir="/root/autodl-tmp/4DGen/exp_data/A_panda_dancing_big_motion" --model_path='/root/autodl-tmp/4DGen/output/2024-08-19/A_panda_dancing_big_motion_21:22:35'

Without modifying the source code and following the provided flow, we get a blurred reconstruction early in the iteration, but an all-white rendering later in the iteration. I need your help, any suggestions are useful for me! Thanks in advance!

[hywang2002]

inputs:

outputs:

settings: python train.py --configs arguments/i2v.py -e A_panda_dancing_big_motion --name_override A_panda_dancing_big_motion

python render_for_eval.py --id="A_panda_dancing_big_motion" --savedir="/root/autodl-tmp/4DGen/exp_data/A_panda_dancing_big_motion" --model_path='/root/autodl-tmp/4DGen/output/2024-08-19/A_panda_dancing_big_motion_21:22:35'

Without modifying the source code and following the provided flow, we get a blurred reconstruction early in the iteration, but an all-white rendering later in the iteration. I need your help, any suggestions are useful for me! Thanks in advance!

Hi, I also encountered this problem, is there any solution?

[YuyangYin]

Hi, thanks for interest. I guess the error is caused by 4DGS, parts of cases fail to reconstrcution. But I also don't know how to solve that.

[YuyangYin]

Hi, thanks for interest. I guess the error is caused by 4DGS, parts of cases fail to reconstrcution. But I also don't know how to solve that.

I find an universal law is if the pseudo lables are not consistent, like the back view of pandas in your cases, the blank results may occur. I think these data is not suitable for 4DGS. And think from a different perspective, if the pseudo labels are not good, the results will be bad, even is not blank. So the performance can be improved by better multi views models.