Parity Partition Coding for Sharp Multi-Label Classification
This work addresses the challenge of multi-label classification in computer vision, offering an incremental improvement with a new metric and coding technique.
The paper tackles the problem of efficiently training and evaluating image classifiers for many object categories by proposing a novel metric called sharpness, defined as the fraction of categories above a threshold accuracy, and introduces parity partition coding to increase sharpness while reducing outputs. The approach outperforms baselines on MultiMNIST and CelebA, requiring fewer parameters and exceeding state-of-the-art accuracy on individual labels.
The problem of efficiently training and evaluating image classifiers that can distinguish between a large number of object categories is considered. A novel metric, sharpness, is proposed which is defined as the fraction of object categories that are above a threshold accuracy. To estimate sharpness (along with a confidence value), a technique called fraction-accurate estimation is introduced which samples categories and samples instances from these categories. In addition, a technique called parity partition coding, a special type of error correcting output code, is introduced, increasing sharpness, while reducing the multi-class problem to a multi-label one with exponentially fewer outputs. We demonstrate that this approach outperforms the baseline model for both MultiMNIST and CelebA, while requiring fewer parameters and exceeding state of the art accuracy on individual labels.