19.08.2010 - TCon
- 1 Minutes of the 19.08.2010 TCon beween CNES and CS
Minutes of the 19.08.2010 TCon beween CNES and CS
- Next release planned for the end of september.
- Keep in mind dashboard watch & repair so as to avoid the dashboard sprint before release.
- New classes should be tested with Tv as well as Tu whenever possible.
- Opening project resources (mailing lists, project discussion ...) to a wider audience has been discussed.
Work in progress
- ImageMetadataInterface has been refactored to distinguish between SAR and Optical images
- A generic SAR calibration filter has been developed
- A patch was submitted to Ossim regarding WV2 metadata reading
- Optical calibration tests have been updated
- The new SAR calibration filter results have been compared to NEST on a TSX image. Looking at NEST source code, a noise bias correction term is not taken into account in NEST, but is in OTB.
- Email NEST developpers to ask why the noise term is discarded. If needed, add a flag to the OTB filter in order to enable/disable this noise term.
- Further validate results with NEST (RMSE, max error).
- Add one more supported sensor: RADARSAT1, and validate with NEST again
- Ensure calibration filters are streamed and multi-threaded
- Update Formosat2 baselines for metadata reading
Mathematical expression parser (muParser)
- Filter has been developed
- Class documentation has been enhanced
- The class name has been changed to otbBandMathImageFilter
- Monteverdi module has been developed
- A user-friendly syntax error handling has been added
- VectorImage import in the module is supported
Object Labeling (Object-based classification)
- Classes from the R&D study have been integrated in OTB "as is" with instantiation tests
- Application has been integrated in Monteverdi as a module
- Some code and functions clean-up have been performed (useless menu, useless methods in model)
Remaining tasks in OTB:
- R&D code integrated in the OTB must be tested for non-regression
- There are still missing filters to implement a full object-based classification pipeline in OTB (conversion between LabelMap and SampleList have to be done by hand as well as normalization for instance). These missing filters have to be developed and the existing one should be enhanced so that a full OBC pipeline can be set-up.
Remaining tasks in Monteverdi module:
- Handling images with any number of bands: the current version assume 4 bands images with NIR and computes radiometric indices like NDVI or NDWI2. We need to have another making no assumption on the image type and computing stats (mean, variance, kurtosis, skewness) on each input image band.
- Handling large scene: This is an open point as we do not have a straightforward solution to do so for now. Besides, streaming with an up-stream mean shift module will be very slow for navigation and practically not useable.
- The module should benefit from new filters developed (model code simplification)
- End-to-end testing has to be set-up for this module.
Vectorization module in Monteverdi
- Module has been made visible from the Monteverdi interface
- Some buttons have been wired
- Properly export the VectorData
- Wire remaining dead buttons
- Load vector datas
- Integrate internship contribution on semi-supervised vectorization
Geometric correction framework refactoring
- New framework prototypes have been developed in OTB-Applications (otbFastOrthoRectif and otbBundleToPerfectSensor)
- A solution for linear VectorImage interpolation has been set-up into the internal ITK
- Apply refactoring as described in the Open issues page
- Update Monteverdi modules to use the new framework
- Filter has been integrated in OTB
- A simple module has been developped for Monteverdi
- Investigate the Nan pixels in output correlation field
New classes will be integrated in OTB-Wrapping for next release. Selected classes are yet to be chosen.