# Counting People by Estimating People Flows

arXiv: https://arxiv.org/abs/2012.00452 · HTML: https://mlsanity.com/paper/2012.00452/

## Key facts

- Cited by 11 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 1 Dec 2020.

## Built on by (most cited first, 11 of 11)

- [Leveraging Self-Supervision for Cross-Domain Crowd Counting](https://mlsanity.com/paper/2103.16291/) (2021)
- [Video Crowd Localization with Multi-focus Gaussian Neighborhood Attention and a Large-Scale Benchmark](https://mlsanity.com/paper/2107.08645/) (2021)
- [Crowd Localization from Gaussian Mixture Scoped Knowledge and Scoped Teacher](https://mlsanity.com/paper/2206.05717/) (2022)
- [Multi-view Tracking Using Weakly Supervised Human Motion Prediction](https://mlsanity.com/paper/2210.10771/) (2022)
- [Consistency-Aware Anchor Pyramid Network for Crowd Localization](https://mlsanity.com/paper/2212.04067/) (2022)
- [A Spatio-Temporal Attentive Network for Video-Based Crowd Counting](https://mlsanity.com/paper/2208.11339/) (2022)
- [Counting Like Human: Anthropoid Crowd Counting on Modeling the Similarity of Objects](https://mlsanity.com/paper/2212.02248/) (2022)
- [Pixel-Wise Symbol Spotting via Progressive Points Location for Parsing CAD Images](https://mlsanity.com/paper/2404.10985/) (2024)
- [Counting with Adaptive Auxiliary Learning](https://mlsanity.com/paper/2203.04061/) (2022)
- [SSR-HEF: Crowd Counting with Multi-Scale Semantic Refining and Hard Example Focusing](https://mlsanity.com/paper/2204.07406/) (2022)
- [Depth-Assisted Network for Indiscernible Marine Object Counting with Adaptive Motion-Differentiated Feature Encoding](https://mlsanity.com/paper/2503.08152/) (2025)

## Abstract (arXiv)

Modern methods for counting people in crowded scenes rely on deep networks to estimate people densities in individual images. As such, only very few take advantage of temporal consistency in video sequences, and those that do only impose weak smoothness constraints across consecutive frames. In this paper, we advocate estimating people flows across image locations between consecutive images and inferring the people densities from these flows instead of directly regressing them. This enables us to impose much stronger constraints encoding the conservation of the number of people. As a result, it significantly boosts performance without requiring a more complex architecture. Furthermore, it allows us to exploit the correlation between people flow and optical flow to further improve the results. We also show that leveraging people conservation constraints in both a spatial and temporal manner makes it possible to train a deep crowd counting model in an active learning setting with much fewer annotations. This significantly reduces the annotation cost while still leading to similar performance to the full supervision case.