SYRAC: Synthesize, Rank, and Count

TL;DR

This paper introduces SYRAC, a novel crowd counting method that leverages synthetic data generated by latent diffusion models to reduce annotation effort and achieve state-of-the-art results in unsupervised crowd counting.

Contribution

It proposes a new approach that uses synthetic data from diffusion models to eliminate manual annotations in crowd counting tasks.

Findings

Achieves state-of-the-art results in unsupervised crowd counting.

Effectively uses ranked image pairs for pre-training.

Utilizes synthetic images with known object counts for improved accuracy.

Abstract

Crowd counting is a critical task in computer vision, with several important applications. However, existing counting methods rely on labor-intensive density map annotations, necessitating the manual localization of each individual pedestrian. While recent efforts have attempted to alleviate the annotation burden through weakly or semi-supervised learning, these approaches fall short of significantly reducing the workload. We propose a novel approach to eliminate the annotation burden by leveraging latent diffusion models to generate synthetic data. However, these models struggle to reliably understand object quantities, leading to noisy annotations when prompted to produce images with a specific quantity of objects. To address this, we use latent diffusion models to create two types of synthetic data: one by removing pedestrians from real images, which generates ranked image pairs with a weak but reliable object quantity signal, and the other by generating synthetic images with a predetermined number of objects, offering a strong but noisy counting signal. Our method utilizes the ranking image pairs for pre-training and then fits a linear layer to the noisy synthetic images using these crowd quantity features. We report state-of-the-art results for unsupervised crowd counting.