– I showed the Dice coefficients for the runs with GMM and SMM. GMM gets better Dice coefficients
and on top of that RGB gives better results than all the other feature vectors tested. From the runs
that I did I also noticed that ConeHSV has the better sensitivity, almost the whole tool is detected,
however, it has some huge fails for particular images.
– I also showed them the poster for the new MRes students.
Comments by Dan:
– The thesis was short on results. Good clinical introduction.
– The superpixels are an interesting idea.
– I should test my algorithm with the MICCAI challenge data available.
– The real-time tracking would work better with TLD rather than with Tomasi features
(this is a pending task since a while ago that I have to test).
– It would be interesting to test my algorithm in videos of fetoscopic procedures without instruments present
to see what happens.
– It would be interesting to test a super-resolution approach for the detection.
Comments by Tom:
– The poster should be with bullet points because it is a visual aid for presenting, not a self-contained document.
– Suggestions on how to improve the wording of the poster.
– Tom is happy that I send the SMM code to the sklearn guys after MICCAI and that I participate in the GSOC 2016
with a GMM related project.