Objectives: investigate improvements to the performance of TM by introduction of a global descriptor. A global descriptor can be, for example, obtained by aggregation of local descriptors. Proposals are strongly encouraged to address the localization problem.
CE conditions:
Usage of TM;
Standard performance evaluation setup on the corrected datasets.
Reduction of vector quantization tables size in TM:
Objectives: investigate low memory solutions for descriptor compression.
The proponents should report the number and dimensionality of tables that need to be stored, type of values stored, and the resulting memory requirements in bytes. Threshold of 5MB per operating point.
SIFT based descriptors retained as a part of the pipeline. Parameters can be modified.
Modification to compression may necessitate change to decompression and matching stages.