Leveraging Self-Supervision for Cross-Domain Crowd Counting

New approach to training on a mixture of synthetic and real data for crowd counting

Released in: Leveraging Self-Supervision for Cross-Domain Crowd Counting



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. The authors remedy this shortcoming by training with both synthetic images, along with their associated labels, and unlabeled real images. To this end, they 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.


Year Released

Key Links & Stats

Cross-Domain Crowd Counting

Leveraging Self-Supervision for Cross-Domain Crowd Counting

@InProceedings{Liu_2022_CVPR, author = {Liu, Weizhe and Durasov, Nikita and Fua, Pascal}, title = {Leveraging Self-Supervision for Cross-Domain Crowd Counting}, booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)}, month = {June}, year = {2022} }

ML Tasks

  1. Object Detection
  2. Scene Understanding

ML Platform

  1. Pytorch


  1. Still Image


  1. Digital Human

CG Platform

  1. Other

Related organizations