The rendering results are very poor, and the entire 3D structure has been reconstructed badly.

[jiangyijin]

I used the dataset from https://www.microsoft.com/en-us/research/project/rgb-d-dataset-7-scenes/ for training and found that the reconstruction effect was very poor. I did not use COLMAP for sparse reconstruction to obtain poses, but instead estimated the poses using my own method. The point cloud input was random. Result:

[jiangyijin]

Can someone tell me why this is happening?