Request DTU dataset

[xuyx55]

Thanks for your 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!

[jiaw-z]

Hello, I have fixed some bugs for running DTU datasets, more details are in #31.

[xuyx55]

Thank you for your reply. I ran the updated code on the scan41 but found that I still could not get a high-quality rendering as shown below. I wonder if this is due to my data processing. And is it possible to release the DTU data if it is convenient?

Novel view rendered by CoR-GS:

Point cloud obtained by COLMAP (3_views/dense/fused.ply)

[xuyx55]

Thank you for your reply. I ran the updated code on the scan41 but found that I still could not get a high-quality rendering as shown below. I wonder if this is due to my data processing. And is it possible to release the DTU data if it is convenient?

Novel view rendered by CoR-GS

Point cloud obtained by COLMAP (3_views/dense/fused.ply)

[jiaw-z]

It is strange to see the rendered images as shown. This is my novel-view rendering by running the updated code

And this is my colmap reconstruction results, which seems similar to yours.

This is the processed data for the 3-view scan41.
https://drive.google.com/file/d/1g1lj3B4tgcSv0tT6avQtcJdrmW9JMCJL/view?usp=sharing

[xuyx55]

I am confused, too. I retried to install the environment in a new virtual environment and ran the code with the supplied MVS point cloud and still got similar results as before. However, CoR-GS works fine with the LLFF dataset, which makes me suspect that there is something wrong with the DTU data. I'd appreciate it if you could release more complete data (including the sparse reconstruction results from COLMAP) and the training log!

[jiaw-z]

Sure, I'd like to share the preprocessed data for DTU, and it may take a few days since I am working on a due.
And maybe you can other scans to see if the code works well.
Or can you provide more training logs such as the visualization results during training and metric results during test iterations.

[xuyx55]

Thanks and looking forward to your sharing. Here's a log of running the code on scan41, metric and visualization results.

Training log

Metric results

Training visualization
500

3000

10000