Tweet
Is there a feature that allows for automasking of the green screen ? Seems a green screen makes more noise than a white background
-
-
Greg -- I found through trial and error that I have to pre-process my images in Photoshop to get a perfect white background. Pre-processing is fairly common for 3D photogrammetry projects.
For a few images, say, 10-20 this can be done manually, where the background is selected, deleted, and the image re-saved. Any reflected glow from the greenscreen on your object can be processed out, too. My object edges became BEAUTIFULLY SMOOTH when I pre-processed my images.
Using and older version of, say, Photoshop 7, which is widely available on torrents (and operating under the assumption that you legitimately had a Photoshop license from back in the day) it is possible to run "batch" edits. You record yourself removing the background using the macro/action editor, then redo that recording for every photo.
I can batch edit several hundred photos at a time this way. It is also possible to use the free software GIMP to batch edit, but the command line process is pretty nerdy and hard to figure out: https://www.gimp.org/tutorials/Basic_Batch/.
-
Greg, if you use a uniform colored background, then the 'turntable mode' + auto-compute in Masquerade should work fairly well. You just need to mask the first image, and then you can autocomplete the other masks (check the 'turntable mode' and click on the cogwheel icon).
Usually, it would be better to use a black or white background. Green screen or colored background can cause color bleeding on the subject.
As for the image preprocessing, you should pay attention. There are specific actions you should always avoid (cropping, or performing lens distortion) and preprocessing filters (like enhance contrast, reduce noise) that some people find useful. Here are my 2 cents. Sfm algorithms are robust to illumination changes, and so many image filters shouldn't do anything at all, while other might degrade the results. However, Since SfM algorithms are statistical, it might also happen to see improved reconstruction sometimes. My suggestion is to not preprocess the images, with the exception where you are dealing with RAW images (as they can contain the sensor noise) or if you just want to improve the texturing in the final phase (you can switch the used pictures before texturing).
Comment
Comment