I’m test driving RC demo mode, intuitive GUI and everything looks great. I’ve read through the FAQ and see very recent threads in the forum regarding the new FBX export feature, have a few questions.
When my project returned an error Out Of Disk Space, I read about the Temp folder loading up with data pretty quickly, see in the project folder that my imported JPGs are copied, but then duplicated again into TIFFs. Are these storage intensive TIFFs necessary to optimizing what’s happening in the SFM engine? If not, is there a preference to stay with JPGs, and if so, why would the imported images even need to be copied over again?
It’s great to learn RC now supports exporting FBX with cameras. My workflow is to bring these cameras into Mari using projectors to paint through to the mesh, which is why I’m less concerned with color at the stage of reconstruction. When I did this coming from **************, I noticed my cameras were never truly aligned to the geometry when viewing the scene through these projectors. An active thread here seems to echo my experience when working with Blender I believe. We know the source photography is slightly distorted, the photogrammetry engine isn’t perfect, will present some drift, so wouldn’t both effects get compounded as you attempt to project from the source imagery back to the end model? Did I read correctly that there’s an option to export the imagery with an FBX that has the same undistort parameters applied as was used in the scene reconstruction? I’m eager to test this.
I’m test driving RC demo mode, intuitive GUI and everything looks great. I’ve read through the FAQ and see very recent threads in the forum regarding the new FBX export feature, have a few questions.
Thanx for kind words…
When my project returned an error Out Of Disk Space, I read about the Temp folder loading up with data pretty quickly, see in the project folder that my imported JPGs are copied, but then duplicated again into TIFFs. Are these storage intensive TIFFs necessary to optimizing what’s happening in the SFM engine? If not, is there a preference to stay with JPGs, and if so, why would the imported images even need to be copied over again?
We do NO conversion of JPGs to TIFFs or etc, that must be there from other source…
CACHE folder -> Yes, we use the cache folder quite intensively so it’s best if you place it on a FAST and LARGE drive and from time to time ( depending on the project size ), clear it in WORKFLOW->SETTINGS->CLEAR CACHE button…
It’s great to learn RC now supports exporting FBX with cameras. My workflow is to bring these cameras into Mari using projectors to paint through to the mesh, which is why I’m less concerned with color at the stage of reconstruction. When I did this coming from **************, I noticed my cameras were never truly aligned to the geometry when viewing the scene through these projectors. An active thread here seems to echo my experience when working with Blender I believe. We know the source photography is slightly distorted, the photogrammetry engine isn’t perfect, will present some drift, so wouldn’t both effects get compounded as you attempt to project from the source imagery back to the end model? Did I read correctly that there’s an option to export the imagery with an FBX that has the same undistort parameters applied as was used in the scene reconstruction? I’m eager to test this.
Yes, you can UNDISTORT it properly on export, all settings can be found in the FBX export settings…
Hello, now driving with a license, eager to work through all the help information.
I have a few questions, first one picking up on where this thread left off. I had changed the Cache Location from System Temp to Project Folder, and yet got suspicious when during reconstruction I got Out Of Disk Space error. The System drive was full, had 46 GB in the Temp folder, but CR showed “Project Folder” in that setting, see attached. I’m aware you have to restart the program for any change to take effect.
I then trashed the images in Temp, closed and opened CR just to be safe, and ran reconstruction again. I didn’t run out of space, but the System Temp folder now shows 16.2 GB (see attached). When I open the Temp folder in C drive it’s empty, is this not the same as what’s indicated in the Storage window? I’m curious to see this folder and verify the images relate to CR.
Secondly, you also addressed the Undistort option when exporting textures, which I’ve now done, but yet to bring all the way to the 3D paint program to see how precisely the projectors paint through to the model. Is the undistort routine based on Exif data coming in from my imagery? Or, does CR produce an undistort based on what it sees in the imagery? From my work in 3D film and testing Zeiss prime lenses to see how closely matched they are, we learned how no two lenses are the same, the variance in fov and such not so important in 2D, but a big deal in 3D. My point is, I don’t quite trust the Exif data one-size-fits-all approach, would seem more promising for accuracy in painting through from cameras if I generate a callibration set specific to my lens/sensor (which won’t change for me). I’m reading about using calibration groups, which would obviously apply in my case, but if I set calibration by exif to True, then this would be the one-size-fits-all approach, my Zeiss 21mm is likely a smidge off from the next. Is there any way to calibrate against my particular lens/body combination?
Lastly, I exported FBX, then OBJ. I wasn’t able to load the former in Maya or Remake. Any thoughts on that?
I have a few questions, first one picking up on where this thread left off. I had changed the Cache Location from System Temp to Project Folder, and yet got suspicious when during reconstruction I got Out Of Disk Space error. The System drive was full, had 46 GB in the Temp folder, but CR showed “Project Folder” in that setting, see attached. I’m aware you have to restart the program for any change to take effect.
I then trashed the images in Temp, closed and opened CR just to be safe, and ran reconstruction again. I didn’t run out of space, but the System Temp folder now shows 16.2 GB (see attached). When I open the Temp folder in C drive it’s empty, is this not the same as what’s indicated in the Storage window? I’m curious to see this folder and verify the images relate to CR.
The best way is to set a CUSTOM folder for RC CACHE, say D:/RC_TEMP, and watch it from time to time ( delete its content in 1-2 days ) But first use WORKFLOW->SETTINGS->CLEAR CACHE button to clear it without searching in folders…
Secondly, you also addressed the Undistort option when exporting textures, which I’ve now done, but yet to bring all the way to the 3D paint program to see how precisely the projectors paint through to the model. Is the undistort routine based on Exif data coming in from my imagery? Or, does CR produce an undistort based on what it sees in the imagery? From my work in 3D film and testing Zeiss prime lenses to see how closely matched they are, we learned how no two lenses are the same, the variance in fov and such not so important in 2D, but a big deal in 3D. My point is, I don’t quite trust the Exif data one-size-fits-all approach, would seem more promising for accuracy in painting through from cameras if I generate a callibration set specific to my lens/sensor (which won’t change for me). I’m reading about using calibration groups, which would obviously apply in my case, but if I set calibration by exif to True, then this would be the one-size-fits-all approach, my Zeiss 21mm is likely a smidge off from the next. Is there any way to calibrate against my particular lens/body combination?
No, we do not use the exif for undistortion as we calculate our own distortion parameters. EXIF is just an informative source of lens parameters.
What settings are you using for the undistortion ?
Lastly, I exported FBX, then OBJ. I wasn’t able to load the former in Maya or Remake. Any thoughts on that?
Hard to say, do you get any warnings or ? isn’t the file too large to load with the texture in the particular apps ?
Best way is to set CUSTOM folder for RC CACHE, say D:/RC_TEMP and watch it from time to time ( delete content after 1-2 days ) But first use WORKFLOW->SETTINGS->CLEAR CACHE button to clear it without looking in folders…
I’ll make this my practice, but I’d like to be deliberate about what’s being cleared from cache, not to make RC do unnecessary repeats. For instance, if alignment has been updated/improved, does clearing cache remove the previous alignment data? That data is small, it’s all the imagery RC generates that needs management, so when does imagery get generated and when in the process is it safe to clear imagery that no longer serves the state of a saved project? You say every two days, but depending on what happens in those two days, that could mean a lot of things.
What settings are you using for the undistortion ?
Undistort.JPG
I used the default. Again, since I’ve not gotten the FBX to open, I’ve also not taken into 3D paint program to test. You’ll advise me if the defaults are appropriate, but I’d like to better understand how the camera calibration might relate here. To optimize calibration, would a) generating calibration group be appropriate in my case with fixed lens/model capture to speed alignment? b) would this be something I should set up once with an ideal set of images containing a registration board placed at various distances and parts of the frame? c) would using this calibration group help optimize what happens at the end with the Undistort on export?
do you get any warnings or etc ? its not too big file + txtr to load them in particular apps ?
No errors reported, but size must be it, +8 GB in apps with only 16 GB RAM. I’ll read up on Simplify, was going to decimate in one of those apps, but whatever works. That said, where would you direct me to learn about a workflow that supports a related set exported in bite-size chunks for various apps? I’ve worked through the section on Components, set control points in overlapping images between various components, nice touch BTW reporting residuals on the fly during placement. With a group of components selected, Align brings all these cameras into a single point cloud, but do those revised orientations ripple back into each component? If so, then I could see how selecting the individual components once again and running reconstruction would then allow exporting the smaller chunks, each properly triangulated against neighboring component meshes and sharing a common scale and coordinate system. Yes? Or is there a different workflow for that?