Hiya,
I’m currently testing out the new v0.7.0 Depthkit version - I can’t seem to work out how to crop the frustum/viewport for individual sensors. I’m trying to remove an obscured section from one of the cameras which is causing an unwanted green colour on the computed model. A point in the right direction would be great!
Thanks so much for posting this issue, and for jumping into our new Pre-release beta! This is extremely helpful for us to see a capture with this issue, it’s something we did not anticipate.
The new version of Depthkit removed the crop feature in order to have it automatically calculate from the bounding box. We hadn’t anticipated that it would also be necessary to crop unwanted texture regions from being included in the blend.
We will discuss internally and see what we can propose for possible fixes.
For future shoots, please keep in mind that you do not need greenscreen as we also removed the Refinement masking feature, and to ensure that there is nothing occluding your capture frames.
Thanks for your response James!
We’re currently using the tape on top of the cameras to avoid lens flare - what are you lot you using to get around that?
We really need a solution in this case for the existing captures though - is there any other way of cropping?
Unfortunately I don’t have a great work around for you just yet. One brutal solution is to remove the offending perspectives from the capture.
We discussed manually cropping the input or output, but that will involve a lot of very precise and error prone frame arithmetic and remapping which is not likely to be workable.
After this latest sprint, we will consider adding crop back to address this specific issue.
Regarding the lens flares, they are definitely an issue that we are constantly fighting.
We have used lens hoods that don’t occlude the color frames. We also often use ring lights that are small & bright, or softboxes that are higher out of the picture.
Other customers have invested in polarization, which can be expensive but work well.
One team shared the following for polarization:
To achieve this effect we have polarised all the lights in our Cyber Rig to the same plane using filter mounts made for our IDA Tronic flashes. Similarly, large polarization sheets can be purchased and cut to cover most light sizes, however similarly they all need to be aligned to the same plane.
We designed and 3D printed some simple filter mounts which slide over the edge of the Kinect (see attached) so that a polarization filter can be cut and held flat over the RGB camera. The tolerance on this cut polarization sheet needs to be tight as the RGB camera is very close to the depth camera and the filter will clip the IR depth data if it covers the sensor. The polarization sheet needs to be placed into the mount over the RGB cam so that it is at 90 degrees to the environment source lighting, thus cross polarizing the texture data.
Let me know how this all sounds
James, please add to your quote the link to the original conversation to look at the 3D print thing, and a link to the mentioned bought filter. It sounds like a good solution even when a polarizer can also change the brightness of the filmed object and cause a micro color shift, but this would be interesting to test.
Greetings Martn
