The problem of the build effect on tum dataset

[Zcl]

Hello, I modified the interface of the input kitti dataset to tum dataset according to the code of orbslam with densesurfelmapping. But I found in the build effect that the actual built map differs a lot from the one in the article, especially when it comes to the f3_office dataset, I don't know what went wrong

While building one side of the table, the point cloud on the other side is gradually decreasing or even disappearing completely
Please give me some guidance if you can, I'd like to build out the tum dataset as it appears in the article