Also- why is it so grainy? I was surprised.
First, I figured out that Zephyr didn't like my equirectangular images because it was 8 fisheye lens cameras instead of the expected 2. Therefore, stitching 8 cameras and THEN chopping up again into the cubemap would not make any sense, too much loss of data.
One solution might be looking for alternate ways to process all of these fisheye images directly.
- Determine lens distortion measurements (maybe Agisoft does this?)
- Unwrap before importing?
- Designate landmarks as a preprocessing step for better alignment.
Another solution might be to simply schedule a reshoot.
- The equipment room has a Ricoh Theta, which is two cameras and a higher resolution.
- Might be worth looking into shooting video instead?
- According to Nick, if shooting with high enough resolution, video can be better data input for photogrammetry software. However, if compressed, model quality is greatly affected (think about MPEGs, they compress with large 'chunks' which would then become stitched). I asked if this was what matchmoving software does and he said no, it's actually different math.
No comments:
Post a Comment