MicMac activity of Marc Pierrot Deseilligny
Sometime, I have difficulty to remember what I did on MicMac last month, year .... so I decided to put some traceabilility. Why on this wiki ? Because, for the future current action, it the opportunity to discuss with other of the priority in MicMac devlopments. For the past, it may help documentation.
Sommaire
curent action
Work on new tie point pipeline, completely free from patents. I am quite optimistic ;-)
Strategic long term action
GRID calibration
When using very fine calibration, all the parametric model (polynom, fourrier ...) may be inapt to modelize high frequence/low amplittude deformation. For this, mesh like model may be more suitable. They used to exist in a very primitive version of MicMac, I mut add them again ...
Finalize Martini
The initial orientation with Martini is operationnal, however the last part (computation of orientation from atomic block) is largely sub-optimal. What has to be done :
- add a hierarchical approach to block construction
- add some structureless bundle adjusment to maintain accuracy of big block orientation
Small thing "to do list"
Big Tiff
Big tiff is now suported by Micmac. It works when reading, but there is some strange result in writing. Now writing is prohibited and the old tiling systems used again. It has to be finalized. A possible solution if the bug is to hard, would be to use the tiling in computation and make a final export in big tiff ?
IGN grid format
The IGN grid format is supported in MicMac, there is few thing to do for them being supported for adjsutment, but is has to be done ...
Past activity 2016
- add accuracy estimation, and other indicator (correlation) in bundle
- work in collaboration with Giang to create new tie points once we have a mesh (hopefully make them more accurate, more mulliplicity, more homogeneously located, less number of tie points)
Past activity 2016
Epipolar 19/09
There was a bug when creating epipolar (is standard mode the degre correction was not inialized to -1)
RPC problem october=
RPC used in satellite modelisation are very instable when used with value outside of their training zone. This create problems when in noisy zone MicMac creates Z outside the validity domain. Modifyed MicMac/Malt so that optionaly the AutoMasq (masq computed of "good" points) reject the points where Z is outside the MinMax.
To come ...
Past activity 2015
To come ...
Past activity 2014
To come ...