Our hardware implementation of kmeans
- Initialization -- pick n pixels at random from the data, then pick k out
of n pixels such that the k pixels are far enough from
one another. Use those k pixels to initialize the k
classes. This is done on the host. The initial mean
table is downloaded to the FPGAs.
- How many classes -- K=16
- Termination -- after J iterations. We are using J=50.
- Distance -- L1 (Manhattan).
This page is maintained
by Prof. Miriam Leeser
Last updated September 16, 1999.
Email: mel@ece.neu.edu