Understanding exr depth File

[khoamaitan]

Hello, I'm working on the latest version of AliceVision,I and want to know more about the depth maps . I'm having problem aligning the point cloud generated from depth map on multi view. As you can see on the two point cloud (attached in this message). However, when I open the corresponding depth map file (exr) in Meshroom (attached in this message), everything is aligned perfectly !!!

The two point cloud visualized in Meshlab

The two depth maps exr files visualized in Meshroom

You can see the difference in the legs.

As far as I know, the exr contains the geometric information about view such as projection matrix, camera intrinsic matrix... These information, I can't access directly by using OpenExr on python. So I use the projection matrix P from the "PrepareDenseScene" node (attached in this message) and the depth value to create the point cloud in the reference coordinates.
I'm wondering what makes such different ? The projection matrix is the combination between the rotation matrix and camera coordination in the SFM file.
Moreover, the min and max in exr header is different from the min and max read by python.

depthMapEXR.zip
pointCloudPLY.zip
PMatrix.zip

I thank you in advance for your help.
Sincerely,
Khoa