About TNT dataset fusion？

[Yin-maple]

Dear Author, I would like to know what method is used for fusion of tnt dataset in your paper?
The results of my run with RunFusion_TAT_Intermediate are quite different from the results in the attachment of your paper, my results contain a lot of ground and trees. Is it because of the missing "blocks"?
I have found that RunFusion_TAT_Intermediate and GeoMVSNet have similar fusion process, is there any correlation between them?
Looking forward to your answers to our questions!
Thank you very much!

[whoiszzj]

Thank you for your attention! Your results contain a lot of ground and trees. I believe this is related to the dataset you used. The dataset we used in our paper is provided by UniMVSNet, which clips the depth_min and depth_max to exclude very peripheral scenes. The "block" is what we used to exclude the sky from the point cloud for visualization, and it has little effect on the actual evaluation results. Our fusion method is also mentioned in the paper, and it is modified based on the point cloud fusion used by UniMVSNet, focusing more on the influence of normals.

[Yin-maple]

Thank you very much for your answer!