[Work Log] Work Log

April 04, 2014

Question: is curve stretching only present in last point? If so, can we hack it away?

  1. detect index stretching (threshold on diff)
  2. for each, visualize with test_optim
  3. fun on full dataset, count (1) how many stretching occur and (2) how many are the last point?


[data, parents] = load_wacv_data(5, [397 530]', params, 1:4:36); %'
[mu, Tracks, params] = get_wacv_result(5, params.model_type, 'deleteme_new4');
out = test_for_wandering_index(Tracks,20);
[wander_is_last, wander_view_num] = exp_2014_04_04_test_wandering_index(Tracks, data, params, 20)

Answer: Usually, wandering is present in last point, but not always. Curve 1 in the example has 15 internal wandering events. All other curves are last point.

Question: Can we correct the bad indices in curve 1 by using a numerical hessian?

Question: could we avoid internal wandering by preventing endpoint wandering? e.g. is internal wandering occurring because endpoint wandering is so extreme?

Question: does stretched point have higher likelihood than hand-picked index? If so, what adjustments to params reverses this case?

Posted by Kyle Simek
blog comments powered by Disqus