Several prototypes of the software required were produced: (i) a very basic opengl rendered to rapidly produce the required images while recording the light direction and camera parameters; (ii) a basic PTM fitter, which is able to produce a reduced feature of a PTM file (does not handle compression, only produces a unique PTM representation LRGB and header information is missing); (iii) a basic Viewer combining the LIght Field with RTI (only uses a simple two plane parameterization and the closes neighbour to approximate the desired ray).

The RTI techniques produces a good result and the implementation in CUDA achieves a good interactive rate. However it was not possible to port the light field evaluation to CUDA. Therefore the generation of a new view is performed is a rate bellow the desirable.

The final image produced produces from the light field has a significant amount of undesirable artefacts mainly due to the nearest neighbour approach to approximate the desired ray. A more robust approach such as bilinear-interpolation or gauss filtering in both planes (s,t) and (u,v) would produce significant improvement in the final image quality.

As an exercise the project was fruitful to consolidate some of the knowledge acquired, mainly in affine transformation, texture interpolation and raytracing.