From: Jim W. <jwa...@ph...> - 2009-09-19 15:49:27
|
Jim Watters wrote: > D M German wrote: > >> Jim> I did run into a problem with the getROI() calculation in PTCommon.c for >> Jim> the fisheye images. The zenith and nadir were both chopped off a bit. >> Jim> I would need to test it but I think if another test was added to do a >> Jim> reverse of a point at the middle of both the zenith and nadir and see if >> Jim> it falls inside the image. If it does then the crop should be expanded >> Jim> to include that area. >> >> What is for any images with FF fisheye or only for those with Tilt? >> > Nothing to do with Tilt. And I suspect even normal lenses might suffer > from the same fate. > I'll investigate. > By adding a couple new test to getROI I am able to test for images that are zenith and nadir and make sure they don't get clipped. I also noticed that on occasion if the panorama is wrapped at the 360 border the ROI would be one or two pixels short of the entire width. This would leave a one or two pixel strip out of the final panorama for that image. By adding another test to see if the ROI is already 95% of the resulting width make it the full width resolves this issue. Using FF fisheye images and 4100X8200 pano I was loosing 100 pixels at the zenith and 45 pixels at the nadir. Making it very obvious something was missing. If I pretended the project had regular images not fisheye then the clipping on the zenith and nadir was reduced. Daniel, If I turn on your debugging code in getROI() to verify the reverse lookup I get all kinds of discrepancies with huge errors for the images with Tilt. I don't think we can ignore this. I find it strange I was able to optimize and get acceptable results. Does this mean that there is something wrong with the tiltForward or tiltInverse functions is math.c ? -- Jim Watters http://photocreations.ca |