Leveraging Self-Supervision for Cross-Domain Crowd Counting

Abstract

State-of-the-art methods for counting people in crowded scenes rely on deep networks to estimate crowd density. While effective, these data-driven approaches rely on large amount of data annotation to achieve good performance, which stops these models from being deployed in emergencies during which data annotation is either too costly or cannot be obtained fast enough.One popular solution is to use synthetic data for training. Unfortunately, due to domain shift, the resulting models generalize poorly on real imagery. We remedy this shortcoming by training with both synthetic images, along with their associated labels, and unlabeled real images. To this end, we force our network to learn perspective-aware features by training it to recognize upside-down real images from regular ones and incorporate into it the ability to predict its own uncertainty so that it can generate useful pseudo labels for fine-tuning purposes. This yields an algorithm that consistently outperforms state-of-the-art cross-domain crowd counting ones without any extra computation at inference time.

Coarse to Fine: Domain Adaptive Crowd Counting via Adversarial Scoring Network

“…To alleviate the issues caused by the domain gap, a technique named unsupervised domain adaptation (UDA) has been preliminarily explored for crowd counting [12,13,29,34]. The key point of UDA is to make use of a domain discriminator to classify patches into the source or the target domains while the deep learner tries to confuse the discriminator in an adversarial manner.…”