On December 25, 2010 02:00:11 pm kfj wrote: > since you refer to panomatic as your reference
this is what milkman used as reference. > , did you take into > account that panomatic also scales down the images? yes - the first two reports are with scaled down images, and I assume they scale both to the same size (since that part has been borrowed from panomatic for cpfind). > To make a valid speed comaprison, you have to also call panomatic with the > --fullscale flag. I'm not into speed comparison at this moment. > > What surprises me is that cpfind's results are worse at full scale than > > resized. > > that is indeed very worrying! > I made a trial, though, and the --fullscale version did take much > longer, but also produced more CPs. with the posted images, or with your own images? we need more data. it is cpfind non-fullscale vs. cpfind fullscale that I was comparing in my last comparison, and that finding was very worrying. > > Please continue discussion and testing there. > > ...oops, saw that too late no worries. for now this is mostly chatter. But we need to collect the body of information available, and it will be a collection over weeks and months and maybe even years and this is where the limits of a mailing list and of human memory are attained. thanks for testing Yuv
signature.asc
Description: This is a digitally signed message part.