Towards Generating Authentic Human-Removed Pictures in Crowded Places Using a Few-Second Video

Area of Science:

• Computer Vision
• Image Processing
• Computational Photography

Background:

• Capturing clear photos of iconic landmarks is often difficult due to crowds.
• Existing methods for removing people can be time-consuming or produce unnatural results.

Purpose of the Study:

• To develop an automated system, Thanos, for generating authentic images of landmarks without people.
• To achieve high-quality results with reasonable computational cost using short video clips.

Main Methods:

• A multi-frame-based recovery region minimization method is proposed.
• Information from multiple image frames is aggregated to minimize the area requiring restoration.
• The system processes short video clips (a few seconds) for efficient operation.

Main Results:

• The proposed method outperforms alternative approaches in generating human-removed images.
• Thanos achieves lower Fréchet Inception Distance (FID) scores compared to existing applications.
• Thanos's FID score was 242.8, outperforming Retouch-photos (249.4) and Samsung object eraser (271.2).

Conclusions:

• Thanos effectively generates authentic, people-free images of crowded landmarks.
• The system offers a computationally efficient solution for image restoration.
• The multi-frame aggregation technique significantly improves the quality of landmark photography.