Efficient GPU-based Decompression of BTF Data Compressed using Multi-Level Vector Quantization

One of the main drawbacks of Bidirectional Texture Function (BTF), as a method of capturing realistic and accurate real-world material appearance, is the resulting size of the measured data set. Several lossy methods to compress the data were proposed over the years to cope with this problem. To efficiently use the compressed data an appropriate decompression algorithms are also needed, allowing fast random synthesis of BTF data without the need to reconstruct the whole BTF back to its original representation. One of such methods based on multi-level vector quantization and providing both good compression ratio and random access from the compressed data was proposed by Havran et al. in 2010. In this paper, we would like to share our experience with writing a GPU based implementation of the decompression part of the aforementioned method. Our goal was to evaluate the implementation difficulty, as well as the resulting performance and suitability of the algorithm for real-time use.