The first export—of a 35 second clip—was fast (less than a minute), but the second one much slower, and the third one took hours (around 12h for a 35 sec clip…). Ideally, we would have had to restart Depthkit after each export.
We lowered the mesh decimation to 1/16, the file-size was reduced but the export took as long. And we could not complete the export for the five cameras.
The OBJ export required 100% of the CPU’s capacity, but very little of the GPU (an Nvidia RTX A4000 card with the latest driver). Is that normal ? Shouldn’t Dephtkit be able to use the GPU ?
Is there any possibility to export all the camera recordings at the same time and not one after the other ?
Is it normal if the OBJ Sequence export of the 1st Kinect (set as the main one) has got less files than the other Kinects for the same recording ? We firstly thought of a data loss but perhaps it’s due to the way your algorithm works.
Would you have any tips to improve our OBJ Sequence export ?
Hi, @OlivierAsselin. I haven’t seen that behavior off successive exports slowing down before, but will try some test exports to see if I can replicate this issue. What color and depth resolution were the 5 sensors set to during the capture you were trying to export?
If the OBJ sequences are different lengths for each sensor, check the underlying recordings, specifically the number of depth frames captured for each sensor in a take. These can be found in the <Project Folder>/<Take Folder>/sensor<#>/depth folders. For a given take, the number of depth frames should be about the same, with the only differences being attributed to dropped frames during capture.
For geometry export workflows, the OBJ export option available in Depthkit is for one sensor at a time, but our upcoming Phase 8 release of our Depthkit Studio Expansion Package for Unity will allow you to take full-quality multiperspective image sequences from Depthkit, reconstruct a 3D asset in Unity, and export a textured PLY sequence for use in other 3D DCC tools. This update will be available in the coming weeks - We’ll be sure to keep you posted.
This textured PLY geometry is also the handoff format of our new 3rd-party integrations with Arcturus’ HoloSuite and Microsoft’s MRCS pipelines, allowing Depthkit Studio assets to be used in any of the publishing platforms supported by those pipelines, including Unreal Engine and WebAR platforms like 8th Wall.
We experienced it with both 1080p and 1440p color resolution.
Both with a depth mode at 640x576
We checked it out and we got the same amount of depth frames captured for each sensor in the take, even though each file doesn’t have the same size. So I guess it’s fine and we can be sure we didn’t go through dropped frames.
And here is a new question :
Do you have a pipeline to export the combined-per-pixel files (or your upcoming method by Multiperspective CPP image Sequence) into Unity to Vuforia for an AR output ?