Crowd counting is a difficult task due to severe occlusion, large-scale variation, and uneven distribution of people. Recently, convolutional neural networks have given hope to solve this problem more easily. However, current approaches require lots of diverse labeled data in the training process.
Hence, a recent paper suggests a method that reduces overfitting and the need for costly labeled data. It uses self-supervised transfer colorization learning. Colorization is generating a color version of a grayscale photograph. The researchers use the idea that the semantics and local texture patterns obtained in the coloration process reflect the density of people in the region. The experiments on several datasets demonstrate that the suggested method achieves better performance given the same labeled dataset as compared with state-of-the-art unlabeled methods.
Labeled crowd scene images are expensive and scarce. To significantly reduce the requirement of the labeled images, we propose ColorCount, a novel CNN-based approach by combining self-supervised transfer colorization learning and global prior classification to leverage the abundantly available unlabeled data. The self-supervised colorization branch learns the semantics and surface texture of the image by using its color components as pseudo labels. The classification branch extracts global group priors by learning correlations among image clusters. Their fused resultant discriminative features (global priors, semantics and textures) provide ample priors for counting, hence significantly reducing the requirement of labeled images. We conduct extensive experiments on four challenging benchmarks. ColorCount achieves much better performance as compared with other unsupervised approaches. Its performance is close to the supervised baseline with substantially less labeled data (10% of the original one).
Research paper: Bai, H., Wen, S., and Chan, S.-H. G., “Crowd Counting by Self-supervised Transfer Colorization Learning and Global Prior Classification”, 2021. Link: https://arxiv.org/abs/2105.09684