Ilya/Results notebook

From Vision Wiki
Jump to navigation Jump to search

Broken Prior Code:

Results file (EM) Iterations (EM) Image list (Py) Exponent for user parameter weighting (ms) for blur (ms) Threshhold on blur (ms) Distance Function (dis) Normal Distribution (EM) Beta Prior (EM) Parameter Range (results) Note
results.mat 1 Full 100 4 .5 Hausdorff 30 Initial run. Noticed that most of the tended to about .9, some to about .6. Probably too much influence of prior.
results2.mat 1 Full 100 4 .5 Hausdorff 16 1 Set to 16 because I wanted the cutoff for where the slope becomes negative for Q to be around dist = 35. Too many annotator parameters end up at either 0, 1, or near .5. I think I do need a prior. The .5s are probably workers that had no assignments, though. I don't even know.
results4.mat 1 Full 100 4 .5 Smoothed Union Intersect () 5 Trying a few different things based on what I saw in the small set. I'm hoping that having a larger data set makes the annotator parameters not all bunch up like they did in the smaller set. But the parameters still ended up all in



Fixed Prior code:

Results file (EM) Iterations (EM) Image list (Py) Exponent for user parameter weighting (ms) for blur (ms) Threshhold on blur (ms) Distance Function (dis) Normal Distribution (EM) Beta Prior (EM) Parameter Range (results) Note
results5.mat 10 Full 100 4 .5 Smoothed Union Intersect 5
results6.mat 10 Full 100 4 .5 Smoothed Union Intersect 15