m24737 – Telecom Italia: presents changes in vector quantization to match the two memory limits established for the CE, 5MB for each operating point, and 128KB overall. Memory limits are matched, with roughly the same performance compared to TM1 for the 5MB case and limited drop for 128KB case. Crosscheck successfully completed (m24857, m24801);
m24759 - Stanford/Nokia: low memory solution ( few KBs) based on SIFT Product lattice quantization. 40/80/160 bits baselines have been defined. Drop in performance (10% at lowest operating points, 2% at higher operating points) compared to TM1. Possible improvements envisaged in applying a Daisy-Lattice coding scheme. Only partial cross-check done so far (m25166).
m24780 – Peking University: A multi-stage vector quantization scheme is presented. Results for both 5MB and 128KB memory limits have been presented. Proposal to consider PSNR as an additional evaluation metric. Two different codebooks are applied, a small codebook as a first step, and a product quantizer as a second step on the residuals. The obtained results show performance comparable to TM1, with a drop at the lowest operating point. The proponents request creation of a CE on interest point selection. Crosschecks have not been completed yet.
