12/10/2004 ___________________________________________ MIAS IRC Oral -Work in progress. http://www.schestowitz.com/Research/Presentations/October_2004/ -Need some appearance model videos for illustration. ___________________________________________ Form 10 -Needs to be completed soon. http://schestowitz.com/Research/Progress/Forms/Year_2/PDF/form10.pdf http://schestowitz.com/Research/Progress/Forms/Year_2/form10.doc ___________________________________________ New Registration Experiments -Epsilon/Delta change. -No systematic experiments yet, but the menu options are in place. -Increasing W_s: There is a new scheme for increasing the weight. -It starts with low values and increases by the power of 4 until reaching 50% roughly. -Intensity models with different shape Weighting and different training set sizes. -3 modes of variation shown in combined model. ___________________________________________ Registration Experiments -See sub-set based registration video. http://www.schestowitz.com/Research/Resources/Experiments/September2004/240904-2/aartlog.htm#see_subset_3_at_precision_e-5_registration (download Video #1 at the bottom) -Need more space on the Computer Science filestore for longer videos to be generated. -Experiments performed: +Increasing resolution by a factor of 10 (100 to 1,000) results in (just under) 50% more time for registration. +Erosion of bumps appears to be slower when image resolution is increased. ___________________________________________ Models at Registration -Intensity accounts for height variation. -Shape accounts for position and width variation. -Combined model depends on shape/intensity weighting and all 3 modes can now be shown in that model, given the correct weighting. -Correct weighting depends on the variation in width, position and height. It may also depend on the size of the training set (Twining). -E-mail messages with examples sent October 8th onwards. ___________________________________________ Integration -Framework in log book. -One way of doing this is clear, only time is needed. ___________________________________________ Landmark Selection -Finding the correct solution: Yet to be addressed unless otherwise decided. -Needs manual annotation and clear understanding of data structures in the code. ___________________________________________ Conclusions -Conclusions were added and described in more detail. http://schestowitz.com/Research/Explanatory_Notes/2004/First_Year_Conclusions/First_Year_Conclusions.pdf ___________________________________________ MATLAB Documentation -Since Ian could not be found, necessary troubleshooting was found in forums. It took no more than an hour to discover (overall time stands at 13 hours). Documentation will be run regularly on the computational server when permission issues are solved. ___________________________________________ Other -Meeting next week needs to be rescheduled because of the plenary meeting. -Venue of Tuesday meetings: Should Angela be asked every week? -Migration of all the research material (thousands of files) and linking is now completed. This is due to project Unity which will change domain names and inevitably break links.