Meeting minutes from the meeting 25 Jan 2017.
What I said:
=======
* New organisation of the upgrade report, particularly of the literature review.
* It would be interesting to make an experiment where we train an FCNs for each available dataset, integrate them afterwards
into a “supernet” and re-train again to hopefully achieve higher generalisation properties. The results will be compared with the
traditional FCN trained with the train data of all the available datasets (obviously without the unseen one). He thinks that this idea
is MICCAI-compatible and that it would be interesting to try it. There is the question of “why this supernet would work better than
just the FCN?”.
What Tom said:
=========
* Move the deformable part modelling to “Geometric models”.
* Move out the convolutional neural networks from the “Classifiers” part.
* Move the “Template matching” to the “Visual representation” section.
* Remove “Model learning” from the title.
* Remove the sentence saying that SVM is non-probabilistic.