OpenMVS_ORB-SLAM3

[ddkats]

Hi @cdcseacave,

I am attempting to use a different pipeline for OpenMVS by substituting the SfM part with Visual SLAM using ORB-SLAM3. I extended the code modifications from docking ORB-SLAM2 to OpenMVS, as proposed here: (https://blog.csdn.net/qq34325730/article/details/118360533). As a result, I obtained the sfm.txt file, which follows this structure:

First, it contains the camera poses with the following columns: ID, fx, fy, cx, cy, qw, qx, qy, qz, tx, ty, tz, imageID.

Next, the file contains the 3D points represented as world positions with the following columns: MapPointWorldPos(x), MapPointWorldPos(y), MapPointWorldPos(z), and observations (indicating the camera poses that observed each point).

Having this txt file along with the images, I added the modified DensifyPointCloud.cpp, considering the read_pose.cpp and read_pose.h files (as suggested here: https://github.com/WangWen-Believer/ORB_SLAM2-to-OpenMVS/tree/main/OpenMVS).

However, the Interface DensifyPointCloud can't recognize the input sfm.txt file and indicates "invalid project." I believe the OpenMVS code has been optimized, and it appears that the DensifyPointCloud interface has undergone some changes in the past two years. Can you provide any help, do you have any suggestions?

[cdcseacave]

the approach is wrong, pls add an interface for your data following https://github.com/cdcseacave/openMVS/wiki/Interface or just imitating one of the many interface in apps

[ddkats]

the approach is wrong, pls add an interface for your data following https://github.com/cdcseacave/openMVS/wiki/Interface or just imitating one of the many interface in apps

Thanks for the quick response. Ok, I see. Based on the above-described input format, do you think it can be used for OpenMVS with a suitable interface for parsing data? Or should I begin with a different approach regarding the input file itself?

[cdcseacave]

I do not know what is the output of Visual SLAM or ORB-SLAM, etc, just check if it is one of the supported formats, of not create a new interface in apps

[longuto]

Docker build is stuck at the position shown in the image, and it has been stuck there for a while.

[hplegend]

the approach is wrong, pls add an interface for your data following https://github.com/cdcseacave/openMVS/wiki/Interface or just imitating one of the many interface in apps

Thanks for the quick response. Ok, I see. Based on the above-described input format, do you think it can be used for OpenMVS with a suitable interface for parsing data? Or should I begin with a different approach regarding the input file itself?

Hi，you have successfuly run dense reconstruction by integrate ORB-SLAM3 and openMVS? I encounter some problem. when I run dense recontruction, there always error just like follwing: "" 01:19:58 [App ] The camera directions mean is unbalanced; the scene will be considered unbounded (no ROI) 01:19:58 [App ] Point-cloud composed of 8016 points with:

• visibility info (35259 views - 4.40 views/point): 106 points with 1- views (1.32%) 2914 points with 2 views (36.35%) 1696 points with 3 views (21.16%) 3300 points with 4+ views (41.17%) 1 min / 4.39858 mean (3.97145 std) / 41 max 01:19:59 [App ] Preparing images for dense reconstruction completed: 143 images (131ms) double free or corruption (!prev) double free or corruption (!prev) ""