I've found only TCL examples for VTK via Cuda. Can you give me advice how I can use VTK via nVidia CUDA with Python?

Implementation of the standardized basic API specification for GIS data handling, "OpenGIS Implementation Specification for Geographic information - Simple feature access" ( http://www.opengeospatial.org/standards/sfa ), would be a great feature to be added to VTK. It should be a killer feature for VTK to be applied to the geographic information area.

Trying to make a real application with vtk that works under windows is nearly imposibble. The problem is that memory allocated by vtk must be consecutive, so it doesn't benefit for virtual memory (swap file memory).
I suggest to create a new class vtkMemory that handle about this. This class stores memory and handle it as it where all consecutive memory, but internally it can be splitted into chunks to allow application to fit into system memory.

As the current import function of QT Creator for cmake projects cannot handle only-header files properly, it would be helpful if a correct QT project file could be generated by cmake similar to Eclipse.

In the same way than matlab is supported, add support to Scilab through adequate classes:
- vtkScilabEngineFilter
- vtkScilabEngineInterface
- vtkScilabMexAdapter

Scilab (http://www.scilab.org/) is an open source program with a very active team. It would be very interesting to benefit both from Scilab and VTK processing functions.

In the end, one may then think to have a Scilab plugin in ParaView to process dataset with Scilab functions, and benefit from ParaView workflow capabilities (filter pipeline) and ability to distribute computations (MPI) to leverage ParaView analysis capabilities.

I have Tested a simple sphere with 0.5 mg polygon with SLI configuration but the performance hasn't chnaged .

What about to make a web version of vtk with webgl?

Help is more for after you exhaust the standard mechanisms. Documentation is a standard and expected resource. We should keep it where it is, but should also add a link to the documentation under resources.

I'd like to be able to build and debug VTK and projects using VTK (Java & C++) in the NetBeans IDE. There's a NetBeans IDE plugin that can interpret Ant projects, but that's an extra step I'd like to avoid.

To make the vtkScalarBarActor more legible over complex scenery, it would help to have a background with programmable color/opacity like the Box object in the legend of the XYPlot actor.
One implementation difficulty will be in creating a box that encompasses the labels & title which hang outside the area defined by the scalarbar actor's position, width & height. The scalar bar widget would need to be aware of this expanded size so that user's could resize the scalar bar by dragging its new edge.

See bug report 9983 for details.

http://www.vtk.org/Bug/view.php?id=9983

When triangulating large volumes, the underlying data type float reaches quickly its limits. Rounding the data to float makes details disappear in case a triangle is far away from the origin of the coordinate system. In case float was choosen in the past for performance reasons, this shouldn't be a reason anymore with modern computers (of course the occupied memory will be twice as much).

Bryn wrote a couple of nice helper functions for dealing with matrices:
http://www.vtk.org/Wiki/Compute_eigenvalues_and_eigenvectors_of_a_symmetric_matrix

These should be wrapped into a class so the interfaces to many of the vtkMath functions (Jacobi(), for example) do not require the user to use ** variables, but rather can easily interface with a matrix using an (r,c) type notation.

Datasets and arrays currently have a vtkInformation object that can be used to store metadata, but none of it is written to disk. Programmers should be able to mark information objects as serializable and all serializable information should be written to disk.

Filters could easily annotate datasets with information that would accelerate processing when reloaded.

Store unit information for each *component* of an array. This could be done easily with vtkInformation objects. For example, the vtkDataArray used by a vtkPoints objects might have 3 vtkInformation objects (one for x, y, and z) specifying the units in which each component's values are specified ('km', 'ft.', 'furlong'). A filter could then perform unit conversions for arrays or even do fancy analysis like using Buckingham Pi theory to determine combinations of arrays to examine.

Provide a way for filter outputs to set or override the parameter values of other filters. For instance, it should be possible (without writing a new class) to have a vtkDescriptiveStatistics filter's mean and median values used as the contour values of a vtkContourFilter.

It would be great if at least the most common filters (stream lines, contour, extract surface...) would take advantage of multiple cores without MPI.