We make available a Java framework that contains a highly optimized implementation of large-scale image search (content-based) based on SURF, VLAD, Product Quantization and Asymmetric Distance Computation, as well as a set of benchmarking utilities that can be used to assess performance trade-offs (between accuracy and memory cost).
lsis.jar contains the implementation of our framework along with extensive developer documentation (Javadoc). There is also user documentation (in pdf format) that includes detailed instructions for reproducing the experimental results of our paper titled: "Exploring performance trade-offs in large-scale image search", currently under review.
We also provide the following compressed data files:
- Pre-computed SURF and SIFT features for the images of the Holidays collection: features.zip
- A selection of pre-computed learning files: learning_files.zip
- A selection of pre-computed indexes: BDB_stores.zip
The instructions for reproducing our experimental results assume that the above compressed files have been extracted in the /data folder of the root folder where our library resides.
A preliminary version of this framework was used to conduct the experimental study in . In case you use this implementation in your research, please cite .
 E. Spyromitros-Xioufis, S. Papadopoulos, I. Kompatsiaris, G. Tsoumakas, I. Vlahavas. An Empirical Study on the Combination of SURF Features with VLAD Vectors for Image Search. In Proceedings of the 13th International Workshop on Image Analysis for Multimedia Interactive Services WIAMIS 2012), Dublin, Ireland, May 2012