Learning Hash Codes via Hamming Distance Targets

Martin Loncaric, Ryan Weber, Bowei Liu

Abstract:

We present a powerful new loss function and training scheme for learning binary hash codes with any differentiable model and similarity function.
Our loss function improves over prior methods by using log likelihood loss on top of an accurate approximation for the probability that two inputs fall within a Hamming distance target.
Our novel training scheme obtains a good estimate of the true gradient by better sampling inputs and evaluating loss terms between all pairs of inputs in each minibatch.
To fully leverage the resulting hashes, we use multi-indexing.
We demonstrate that these techniques provide large improvements to a similarity search tasks.
We report the best results to date on competitive information retrieval tasks for Imagenet and SIFT 1M, improving recall from 73% to 85% and reducing query cost by a factor of 2-8, respectively.

Keywords: information retrieval, learning to hash, cbir

TL;DR: We present a new loss function for training any differentiable model to hash that can vastly improve recall and lookup speed.