We’re testing different cameras to link with the sensors but always get stuck at the camera pairing step…
The lens profile seems good, the calibration chart is well lit, the pink dots are always visible as we move the chart around and the chart stays within the camera view.
We are currently testing the cinema capture with a black magic ursa mini pro paired with a tokina 11-20mm 2.8.
The sensor is also as close as it can be to the camera lens.
As written in the documentation, a minimum of 3 sample is required, but we tried with 20 sample, as well as up to 90 and never got a satisfying result…
We would love to know more about the positioning of the chart as it seems to be our main issue here.
Should we move the chart and keep the camera/sensor stable, or its best to move the camera/sensor and only move the chart forward and backward?
Thank you !
Hi, @PatriceLaCroix! Sorry to hear you’re having trouble getting a good pairing.
I usually get around 20-30 samples per pairing, which should be plenty. Moving the chart rather than moving the camera helps ensure that the position of the sensor doesn’t slip relative to the camera, but if its properly secured, then you can carefully move the camera to get the chart in different parts of the volume.
For every sample, make sure the chart is directly facing the camera/sensor. If the chart is panned/tilted away, the quality of the sample will be lower.
Are you able to share some screenshots of what you’re seeing in the Depthkit interface that indicates poor results? Misaligned color & geometry in the 3D Viewport with the Pairing Preview Enabled? Poor metrics? Let us know, and we may be able to narrow down the issue.
Here’s a screenshot of the interface, and on the second picture is the different placements of the chart.
For each position, I captured different angles by moving the head of the tripod (as seen on the YouTube tutorial).
I just tried a new pairing, moving the chart and head of the tripod around but making sure the chart is always facing the lens (keeping the chart in the center of the frame) but the results aren’t any better… Really confused here !
Would it then be the gap between the camera and the sensor?
The body of the camera makes it impossible the get it closer…
Hi, @PatriceLaCroix - Thanks for sharing these. From what I can see, the calibration doesn’t look bad in the 3D viewport, but the metrics are low.
To clarify about the angle of the chart and the camera, the chart should always face the camera directly, but the camera can pan and tilt so that it’s not always pointing directly at the chart. In other words, the camera should always see the chart facing it squarely even if it’s in different parts of the camera’s frame.
I noticed in both of these calibrations that the Sampled Volume metric is low. This metric measures how many different unique positions in front of the camera have been sampled. What might help is to, in addition to getting the sample positions you are capturing, capture some samples at an even greater distance. Something that helps me fill up the Sampled Volume is to imagine that the camera’s field of view is a solid pyramid extending out from the sensor, and each sample is carving out a small section of that pyramid. Each sample should carve out a new portion of the pyramid; If you get an additional sample in the same part of the volume as a previous sample, it’s not contributing much to the calibration.
Thanks again for helping us make things work ! I had a very successful pairing yesterday and wanted to redo the whole thing just to double check the workflow, and strange enough the metrics are very good but the pairing accuracy is at 0%…
Is this a common software issue or it’s the pairing that’s actually bad (I would be surprised considering the metrics) ?
Have you made a test recording to validate this pairing? If it aligns well, then it looks like there may be an issue with the way Depthkit is reporting the Accuracy metric. Can you zip up and share this project file with us? I’ll have our team take a look to see if we can figure out what’s going on.
Metrics aside, if the pairing is working well as far as texture alignment with your capture, feel free to continue to use it!
Hi again ,
After trying different techniques, we finally found how to get a very good pairing accuracy (yay!) , but noticed that it was only working when our camera is set to 4K… On the documentation it says that we should be able to record up to 8k but after trying 6k and 4.6k we noticed that the camera resolution was the issue. Do you have an idea on how to solve this problem?
Hi, Patrice. When using the full resolution in camera, can you confirm the resolution of the following:
- The images fed into the Lens Profile Creator to generate the lens profile.
- The sample videos fed into Depthkit’s Camera Pairing Stage
- The videos paired to your Depthkit recordings
- The resolution that the camera is set to when generating materials for the above steps. (It should be uniform for all of them.)