search

CUHK-AIM-Group / EndoGaussian

EndoGaussian: Real-time Gaussian Splatting for Dynamic Endoscopic Scene Reconstruction
https://yifliu3.github.io/EndoGaussian/
MIT License
100 stars 5 forks source link

the effect on the Scaled dataset #22

Closed wwj-53 closed 2 months ago

wwj-53 commented 3 months ago

Hello, I used keyfream1 from the scaled dataset to run the code. The dataset has been processed as required. I ran as follows: python train.py -s /media/wwj/3dGSDATA/dataset_3/keyframe_1 --port 6017 --expname dataset_3/keyframe_1 --configs arguments/scared/d3k1.py 。 Then Python render. py -- model_path output/dataset_1/keyframe_1-- skip_train -- skip_video -- configs arguments/scaled/d3k1. py

The results obtained are as follows： Metric evaluation progress: 100%| 11/11 [00:00<00:00, 12.03it/s] Scene: output/try_scared SSIM : 0.6458519 Scene: output/try_scared PSNR : 20.9398842 Scene: output/try_scared LPIPS: 0.4690880 Scene: output/try_scared RMSE: 51.9106958

The rendered image is shown below： 00000

May I ask how I can adjust to achieve the effect close to the paper? (I used the parameters of your code and the Endonerf dataset to achieve the effect close to the paper, but the effect on the Scaled dataset is very poor)?

yifliu3 commented 2 months ago

Hi, thanks for your attention.

We haven't found similar problems before, and I think something unusual happens. To find the problem, can you provide more information, like:

  1. Can other subdatasets normally trained and rendered?
  2. Have you pre-processed the dataset following README?