I don't know why the point cloud checked by the DSOviewer is different from the point cloud of the saved output ply file.

[beepbeep08]

I don't know why the point clouds checked by the DSOviewer and the point cloud of the saved output ply file is different.
the point cloud checked by the DSOviewer is perfect, but output ply file is not.
It even seemed that only two points were saved.The number of lines stored and cnt_point(kf->numPoints()) is the same. and I checked output ply file through the point cloud viewer tool Meshlab .
Why is all the data stored so differently? my data have loop closing.The number of lines stored ply file and cnt_point is the same.

please help me.

[mgladkova]

Hi @beepbeep08. Have I understood you correctly that you are referring to the differences between a pointcloud visualized by DSOViewer and a pointcloud saved in PLY file? If so, please note that DSOViewer filters points with high depth variance for visualization (see arguments in refreshPC function call). I presume if you apply the same thresholds while saving PLY file, your pointclouds would be the same.

[beepbeep08]

Hello,@mgladkova. Thanks for answer. I apply the same thresholds while saving PLY file, but Nothing has changed.
PLY file have a 400,000 cnt_point, meshlab visualized only 3~10 point.
run your code with kitti dataset, A pointcloud visualized by DSOViewer and a pointcloud saved in PLY file are same.

What else can I do?

[mgladkova]

That's quite strange, indeed, but we can try to debug it together.

1. When do you call the function to save a pointcloud? In case you need an example, there is one for TUM Mono sequences.
2. Also can you check you .ply file, in particular value of element vertex entry?

[beepbeep08]

1.i call the function to save a pointcloud at same section like one for TUM Mono sequences.
and also I call the function in viewer's run() function, before " printf("QUIT Pangolin thread!\n");" line. both saved files is exactly same.

2. I can check my ply files, the entrys are like this.
   ply
   format ascii 1.0
   element vertex 76668
   property float x
   property float y
   property float z
   end_header
   -1.06285 -0.661688 0.84488
   -0.615564 -0.662488 0.845402
   -0.517721 -0.662663 0.845516
   -0.18226 -0.663263 0.845908
   0.1532 -0.663863 0.8463

3. I try to visualized the ply file in Unity, and I could see that the stored pointcloud data's central axis strangely, but the details were stored somewhat similar. if you want my ply file, example ply file is here. to visualize in Unity, the file is converted acsii to binary little-endian. this file is original file.
   map_out_final.zip
   it about the 3m length*1m width hallway.

[mgladkova]

Thank you for the information! I have tried to visualize your map_out_final.ply file in meshlab and I can see a pointcloud of more than 10 points (meshlab reports 76668 vertices). The coordinate system of 3D points is local to the first camera pose, since the latter is assigned to an identity transformation. Nevertheless, in meshlab you can transform the pointcloud using Manipulations tool, which includes rotation, translation and scale of your choice.

[wz8530]

Has anyone solved this problem?

[Richardcvpr]

感谢您的信息！我尝试在 meshlab 中可视化您的 map_out_final.ply 文件，我可以看到超过 10 个点的点云（meshlab 报告 76668 个顶点）。3D 点的坐标系对于第一个相机位姿而言是局部的，因为后者被分配给恒等变换。尽管如此，在网格实验室中，您可以使用操作工具来转换点云，其中包括您选择的旋转、平移和缩放。

is it possible to solve this problem by modify the code ? or how could i solve this problem?