Timezone: »

 
Poster
Learning to Agglomerate Superpixel Hierarchies
Viren Jain · Srinivas C Turaga · K Briggman · Moritz N Helmstaedter · Winfried Denk · H. Sebastian Seung

Wed Dec 14 08:45 AM -- 02:59 PM (PST) @ None #None

An agglomerative clustering algorithm merges the most similar pair of clusters at every iteration. The function that evaluates similarity is traditionally hand- designed, but there has been recent interest in supervised or semisupervised settings in which ground-truth clustered data is available for training. Here we show how to train a similarity function by regarding it as the action-value function of a reinforcement learning problem. We apply this general method to segment images by clustering superpixels, an application that we call Learning to Agglomerate Superpixel Hierarchies (LASH). When applied to a challenging dataset of brain images from serial electron microscopy, LASH dramatically improved segmentation accuracy when clustering supervoxels generated by state of the boundary detection algorithms. The naive strategy of directly training only supervoxel similarities and applying single linkage clustering produced less improvement.

Author Information

Viren Jain (Howard Hughes Medical Institute)
Srini C Turaga (Howard Hughes Medical Institute, Janelia Research Campus)
K Briggman (MPI)
Moritz N Helmstaedter (Max Planck Institute for Medical Research)
Winfried Denk (Max Planck Institute for Medical Research)
H. Sebastian Seung (Massachusetts Institute of Technology)

More from the Same Authors