Hi @JyLeev
What were your alignment settings? Have you set the camera positions as fixed during alignment?
What were your XMP export settings?
After importing XMPs into new project, were the images’ positions fixed?
What do you mean by transform was identical? If the XMPs were set to fixed, there shouldn’t be any movement. Were the calibration poses same as prior poses?
What kind of export have you done?
Hi. @OndrejTrhan
Here are the answers for your question.
Q1. What were your alignment settings? Have you set the camera positions as fixed during alignment?
A1. Yes. All cameras are fixed on frame, and not moved or touched during the capture.
Q2. What were your XMP export settings?
A2. I used default setup. We run RC with CLI or .rccmd file. We used exportXMP command without input.
Q3. After importing XMPs into new project, were the images’ positions fixed?
A3. Yes. As mentioned on A1, cameras are fixned on frame.
Q4. What do you mean by transform was identical?
A4. We have 2 set of project. one is to align and get xmp files, and the others reuses the xmp files.
Command “addImageWithCalibration” was used when reuse xmp files.
Therefore, the infomation of each image was same, the align project and the others. That is what ‘identical’ means.
If the XMPs were set to fixed, there shouldn’t be any movement.
→ This is what I’m talking about. why it doesn’t on our case?
what we tried…
inpPosePriorRelative=0~2(Unknown, Draft, Exact)
inpPose=3(Locked)
inpCalibration=2(Fixed)
Q5. Were the calibration poses same as prior poses?
A5. Yes.
Q6. What kind of export have you done? -
A6. Please check xml files. the xml files that we used to export FBX, OBJ files. X4T_FBX.xml (3.1 KB) X4T_OBJ.xml (2.3 KB)
Q1. What were your alignment settings? Have you set the camera positions as fixed during alignment?
A1. Yes. All cameras are fixed on frame, and not moved or touched during the capture. I meant if the position for the images in the RealityCapture were set as Fixed, not the physical cameras in the rig.
Q2. What were your XMP export settings?
A2. I used default setup. We run RC with CLI or .rccmd file. We used exportXMP command without input. The pre-defined settings for export XMPs are set as Draft. So, they can be moved in the new project. If you want to keep the positions, you need to use Export as locked setting.
Q3. After importing XMPs into new project, were the images’ positions fixed?
A3. Yes. As mentioned on A1, cameras are fixned on frame. Here I meant also the Prior pose settings, not physical cameras.
Have you used “addImageWithCalibration” for each image?
If the XMPs were set to fixed, there shouldn’t be any movement.
→ This is what I’m talking about. why it doesn’t on our case? I think this is because (if you used just exportXMP command) the XMPs are set as draft.
Q5. Were the calibration poses same as prior poses?
A5. Yes. Have you checked this in 1Ds view?
However, upon export, I found that some frames were exported with a translation which I never edited applied. I checked you settings. By translation, do you mean exported images with model? If so, this is because you are exporting undistorted images with settings which are casing this. But basically, all undistorted images will be different to the original one. Do you need these undistorted images?
I meant if the position for the images in the RealityCapture were set as Fixed, not the physical cameras in the rig.
→ I tried to set it fix when align and it does not work.
*What were your XMP export settings?
→ I align images without position fix, and make xmp files with Locked option. Option used… X4T_XMP.xml (321 Bytes)
→ Other frames used those xmp files.
Have you used “addImageWithCalibration” for each image?
→ Every image files imported with addImageWithCalibration command with matching XMP file.
Here I meant also the Prior pose settings, not physical cameras.
→ this time tried these options.
used “lockPoseForContinue” after algin and before export
I think this is because (if you used just exportXMP command) the XMPs are set as draft.
→ Problem still exist.
I checked you settings. By translation, do you mean exported images with model? If so, this is because you are exporting undistorted images with settings which are casing this. But basically, all undistorted images will be different to the original one.
→ What I asked is about generated mesh, not image. images are fine for now.
here is the RCCMD files that we generated and used. Pathe are claasified. ALING.rccmd (314 Bytes) 0001.rccmd (2.7 KB)
Can you show also your result and what you consider as wrong to make it more clear?
I also checked your commands and there are some unnecessary settings.
As you are exporting the XMPs as Locked (according to the X4T_XMP.xml file), you don’t need to use these commands in the 0001.recmd file: -editInputSelection "inpCalibration=2" -editInputSelection "inpPose=3" -editInputSelection "inpPosePriorRelative=2"^
Also, I am not sure if this command is necessary in the ALIGN.rccmd file: -editInputSelection "inpCalibration=2"^
Are your data already calibrated to use the fixed calibration for first alignment?
The problem I discovered was that Reality Capture translate the center of mesh as its origin during export. However, the center kept changing due to noise meshes.
So, I tried changing the export option for the coordinate system to “same as XMP” and obtained the desired result.
Additionally, I confirmed that the result comes out without any issues when excluding the unnecessary options you mentioned.
Our images contain focal length information in the EXIF tags, and to use that information, we fixed the calibration values. When we first discovered the mesh translation issue, we attempted to match our data with the data generated by Reality Capture, which led us to try fixing that value.
