This was quite a busy day.
First we meet that guy from UCLA who was presenting his work on DNA sequences. He had programmed in python a software that could handle dan (sub) sequences, compare them and do many other neat features. Thanks to objet-oriented design of python everything was treated more or less like a string, which allowed to handle lot of things like small sequences research in bigger sequences, or even in mltiple genomes through one operation. The beauty of the thing, on top of being so weel intergated in python, is to be able to access online remote, possibly distributed, databases for storage of the genomes. Very interesting on the software design point of view, this teacher also give me several insight on database issues. According to him, MySql was too slow for his work. We have another issue as we handle huge images datasets.
Later, I went visiting M. Desbruns, one of the world leader in discrete differential geometry. I had invited him to my interview the first time I came to Caltech because I wanted to work with him on several unfinished projects of mine.
I have unfisnished work on the problem of "shortest loops on manifold". A paper was accepted at the IEEE EMB annual conference in shanghai in 2006, but the final paper is still to be written. I need to finish the code (but QE is in the way ..) and to compile the results. All the theory and the code design is done, there is no more scientific challenge in this (well, we still need to argue about the isotropy of the sampling of the surface, i.e. the regularity of the mesh to bound the complexity of the algorithm. We also should present a geodesic version instead of a discrete version which is now to fail in some case.
I would also finish my PhD work on WBBM framework for Active surfaces. This need a complete rewriting of a wavelet based analys of 2-manifolds and then of my approach. Of course, a simpler way would be to use my oldest code to get just enough results to publish, but that's kind of against everything I fought for the past years.
Finally, we know that segmenting on 2D (getting contours of cells in datasets) then trying to reconstruct the object in 3D from a stack of contour is a doomed approach. Working directly in 3D already gives better results as you take into account the inter-slices continuity you lose when working slice by slice. We also came to the conclusion that more than 2-manifolds, 3-manifolds are the objects we are looking for. We are dreaming (with eric) of a nice datasrtucture that would handle 3-manifolds and their boundaries (2manifolds) transparently. It's quite a challenge as you loose all the nice orientation information you had on the surface when you go inside a 3-manifold. But if we could implement boundary operator, star operator and other methods of 3-manifolds, lots of work from abstract mathematic could be directly imported for our work. Still a dream.
Desbruns was interested in the first and last topics. We should take contact a little bit later to start the real work as we're both busy for the time being.