The files provide an implementation for our paper:
Stefan Schubert, Peer Neubert, Peter Protzel (2021): Fast and Memory Efficient Graph Optimization via ICM for Visual Place Recognition. In Proc. of Robotics: Science and Systems (RSS).

Run demo.m to replicate the result for StLucia 100909_0845 vs 190809_0845 from our paper.
The provided test data contains the already precomputed pairwise image similarities S, intra-database similarities S^DB, intra-query similarities S^Q and the ground truth with NetVLAD.
While the computation of ICM_mul is quite fast, ICM_min takes some minutes. The memory usage is low in both cases.

After running the code, the following lines should be printed in your Matlab console:
>> demo
Area under curve (AUC) for StLucia 100909_0845 vs 190809_0845:
Raw: AUC=0.41
ICM_mul (DB): AUC=0.46
ICM_min (DB): AUC=0.51
ICM_mul (DB+Q): AUC=0.5
ICM_min (DB+Q): AUC=0.57
ICM_mul (DB+Q+Seq): AUC=0.8
ICM_min (DB+Q+Seq): AUC=0.87
SeqConv (Seq): AUC=0.76



If there are any questions, please feel free to contact:
Stefan Schubert (stefan.schubert@etit.tu-chemnitz.de)

