search

Fictionarry / DNGaussian

[CVPR'24] DNGaussian: Optimizing Sparse-View 3D Gaussian Radiance Fields with Global-Local Depth Normalization
https://fictionarry.github.io/DNGaussian/
Other
246 stars 15 forks source link

Request DTU dataset #35

Open Brucess opened 1 month ago

Brucess commented 1 month ago

Very outstanding and wonderful work! May I ask if it is possible to provide the DTU dataset used in the paper, as the performance has decreased after trying colmap DTU myself. Thank you a lot!

Fictionarry commented 1 month ago

Hi, the DTU dataset we used is preprocessed in the same way as our instruction. What is the detailed phenomenon of the performance decrease? Are there some scenes that become bad, or is the overall quality?

Acturally, with the update of the rasterization module (https://github.com/Fictionarry/DNGaussian/commit/14c3c6a41a32d5182801c11dfd99515b76281b7c), some performance differences appeared on the DTU dataset compared to our released checkpoint, such as the performance on scene21 and scene40 decreased, while on some other scenes increased. After our check and slight adjustment, the overall performance can still reach and even exceed the reported. Thus, it's right if only the metrics on some scenes are different from our released. I'm not sure if it is the problem you meet but hope the explanation can help you.

Brucess commented 1 month ago

After using my colmap DTU dataset for training, I found that the average PSNR of 15 scenarios was lower than that reported in the paper about 1 PSNR. I did not make any changes to other parts and follow the provided instruction of DTU dataset processing, so I suspect that the quality of my colmap DTU dataset has deteriorated. I really hope you can provide your DTU dataset!

Fictionarry commented 1 month ago

After using my colmap DTU dataset for training, I found that the average PSNR of 15 scenarios was lower than that reported in the paper about 1 PSNR. I did not make any changes to other parts and follow the provided instruction of DTU dataset processing, so I suspect that the quality of my colmap DTU dataset has deteriorated. I really hope you can provide your DTU dataset!

Seems there are some problems. Contact me by email and I'll send you the preprocessed dataset.