Currently all the results from the color correlogram tool are serialized in a feature vector per file way. This causes the the second tool in the chain (FeatureMatcher) to take a long time deserializing (90% of execution time when run on 70k files).
Test if single file serialization improves performance