ANTHROPOS-V: benchmarking the novel task of Crowd Volume Estimation

TL;DR

This work defines Crowd Volume Estimation (CVE) as predicting the total body volume occupied by crowds from RGB images and introduces ANTHROPOS-V, a synthetic, photorealistic dataset with per-person and per-part volume annotations. It proposes Per-Part Volume Density Maps and the STEERER-V model, achieving state-of-the-art performance on ANTHROPOS-V and robust transfer to real-world datasets, outperforming baselines adapted from Crowd Counting and Human Mesh Recovery. The study provides a formal CVE objective $\min_{\theta} \lVert V_{tot} - M_{\theta}(I) \rVert$ and CVE-specific metrics MAE and PP-MAE, along with a comprehensive benchmark including data generation, SMPL fitting, and real-world transfer experiments. The results highlight the necessity of CVE-tailored supervision over naive counting strategies and underscore the practical impact for crowd management, safety, and infrastructure assessment, while acknowledging ethical and bias considerations for deployment.

Abstract

We introduce the novel task of Crowd Volume Estimation (CVE), defined as the process of estimating the collective body volume of crowds using only RGB images. Besides event management and public safety, CVE can be instrumental in approximating body weight, unlocking weight sensitive applications such as infrastructure stress assessment, and assuring even weight balance. We propose the first benchmark for CVE, comprising ANTHROPOS-V, a synthetic photorealistic video dataset featuring crowds in diverse urban environments. Its annotations include each person's volume, SMPL shape parameters, and keypoints. Also, we explore metrics pertinent to CVE, define baseline models adapted from Human Mesh Recovery and Crowd Counting domains, and propose a CVE specific methodology that surpasses baselines. Although synthetic, the weights and heights of individuals are aligned with the real-world population distribution across genders, and they transfer to the downstream task of CVE from real images. Benchmark and code are available at github.com/colloroneluca/Crowd-Volume-Estimation.

Figures (4)

• Figure 1: ANTHROPOS-V is the first dataset for the novel task of Crowd Volume Estimation. It features human crowds engaged in activities within real-world environments ( left). Each individual in the dataset is labeled with ground-truth SMPL shape parameters ( center) along with their volume labels. We employ novel Per-Part Volume Density Maps as a superior supervision signal for training models to address this task ( right).
• Figure 2: STEERER-V's PP-MAE$/$MAE computed on ANTHROPOS-V test set.
• Figure 3: Examples from ANTHROPOS-V, showcasing several lighting and weather conditions, camera angles, and a variety of physiques. The crops in the zoomed boxes depict persons with differences in statures and body shapes.
• Figure 4: Statistical analysis of the distributions of mass, height, and Body Mass Index (BMI) of the individuals in ANTHROPOS-V. Solid curves depict the empirical distributions, while dashed curves refer to the theoretically expected ones silverman2022exact. (\ref{['subfig:original_stats']}) Distribution of the body features of the characters' meshes in GTA-V without any manipulation. (\ref{['subfig:modified_stats']}) Distribution of the body features of the characters' meshes in GTA-V after applying some geometrical transformations. (\ref{['subfig:smpl_stats']}) Distribution of the body features of the resulting fitted SMPL meshes.