Reducing Spatial Labeling Redundancy for Active Semi-Supervised Crowd Counting.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Crowd counting requires extensive labeling, making semi-supervised methods crucial for reducing annotation burden.
• Existing semi-supervised methods often use a suboptimal None-or-All labeling strategy, focusing on similar individuals within selected images.

Purpose of the Study:

• To reduce spatial labeling redundancy in semi-supervised crowd counting.
• To develop a more efficient and effective labeling strategy for crowd counting tasks.

Main Methods:

• Proposed a method to annotate only representative regions, analyzing representativeness from density maps.
• Formulated region representativeness using cluster centers of Gaussian Mixture Models.
• Leveraged feature propagation to supervise unlabeled regions, exploiting similarities among individuals.

Main Results:

• The proposed method significantly outperforms previous state-of-the-art approaches on widely-used benchmarks.
• Demonstrated the effectiveness of annotating representative regions and using feature propagation for improved crowd counting accuracy.

Conclusions:

• The developed method offers a more efficient and accurate solution for semi-supervised crowd counting.
• Breaking the traditional labeling chain and reducing spatial redundancy leads to substantial performance gains.