Alert button

Deep Momentum Uncertainty Hashing

Sep 17, 2020
Figure 1 for Deep Momentum Uncertainty Hashing
Figure 2 for Deep Momentum Uncertainty Hashing
Figure 3 for Deep Momentum Uncertainty Hashing
Figure 4 for Deep Momentum Uncertainty Hashing

Share this with someone who'll enjoy it:

Discrete optimization is one of the most intractable problems in deep hashing. Previous methods usually mitigate this problem by binary approximation, substituting binary codes for real-values via activation functions or regularizations. However, such approximation leads to uncertainty between real-values and binary ones, degrading retrieval performance. In this paper, we propose a novel Deep Momentum Uncertainty Hashing (DMUH). It explicitly estimates the uncertainty during training and leverages the uncertainty information to guide the approximation process. Specifically, we model \emph{bit-level uncertainty} via measuring the discrepancy between the output of a hashing network and that of a momentum-updated network. The discrepancy of each bit indicates the uncertainty of the hashing network to the approximate output of that bit. Meanwhile, the mean discrepancy of all bits in a hashing code can be regarded as \emph{image-level uncertainty}. It embodies the uncertainty of the hashing network to the corresponding input image. The hashing bit and the image with higher uncertainty are paid more attention during optimization. To the best of our knowledge, this is the first work to study the uncertainty in hashing bits. Extensive experiments are conducted on four datasets to verify the superiority of our method, including CIFAR-10, NUS-WIDE, MS-COCO, and a million-scale dataset Clothing1M. Our method achieves best performance on all datasets and surpasses existing state-of-the-arts by a large margin, especially on Clothing1M.

Share this with someone who'll enjoy it: