Confusing results for DTU dataset with known official pose

[Gaozihui]

Hi, I use DTU official pose as known prior to get colmap results by triangulation with known camera posesand then put it into gs for rendering. However, I found the rendering images are not good as original colmap resutls( PSNR now 29 VS original 35) and the image is a little bit blurred. Also the convergence now takes longer time with worse results. Here i give an example: I train the gs for 10k iteration but the image is still blurred.

[OiOchai]

Hi, I use DTU official pose as known prior to get colmap results by triangulation with known camera posesand then put it into gs for rendering. However, I found the rendering images are not good as original colmap resutls( PSNR now 29 VS original 35) and the image is a little bit blurred. Also the convergence now takes longer time with worse results. Here i give an example: I train the gs for 10k iteration but the image is still blurred.

I have the exact same problem that GT works worse than colmap! https://github.com/hustvl/4DGaussians/issues/100 Have you found the solutions?

[Wuuu3511]

Hi, I use DTU official pose as known prior to get colmap results by triangulation with known camera posesand then put it into gs for rendering. However, I found the rendering images are not good as original colmap resutls( PSNR now 29 VS original 35) and the image is a little bit blurred. Also the convergence now takes longer time with worse results. Here i give an example: I train the gs for 10k iteration but the image is still blurred.

Did you solve this problem?