Late-hour realization: not sure how to handle attachment matrix when multiple-views are allowed. Deriving the correct equations will take all afternoon; no time. Decided to remove the tracking part from the paper.
code in wacv-2012/run_wacv.m
. Example call:
gt_paths = arrayfun(@(x) sprintf('wacv-2012/datasets/%d', x), 1:11, 'UniformOutput', false);
gt_cam_fmt = [gt_paths{i} '/cal/calib_%d.txt'];
gt_fname = [gt_paths{i} '/ground_truth_2d.gt2'];
[mu, lengths] = run_wacv(gt_fname, gt_cam_fmt, 1:4:36, [397,530]', params, 1);
mu = reshape(mu, 3, []);
mu = mat2cell(mu, 3, lengths);
colors = lines(numel(mu));
for i = 1:numel(mu);
c = mu{i};
col = colors(i, :);
plot3(c(1,:), c(2,:), c(3,:), '-o', 'Color', col);
hold on
end
Dataset #1: curve #1 has some weird shape. Probably and error in GT
Dataset #2: again weirdness in curve #1. Now I'm guessing it's an attachment issue:
Run again without attachments.
Dataset #2: curve 1 issue is better; now exhibiting some weird stray points. few views, bad indices?
It looks like the base curves are trying to attach to the end of curve #1, instead of the beginning.
Posted by Kyle Simek