Zero-Shot Scene Change Detection
This work addresses scene change detection for applications like surveillance or robotics, offering a domain-agnostic solution, though it is incremental as it builds on existing tracking models.
The authors tackled scene change detection by introducing a training-free method that adapts tracking models to detect changes between reference and query images, addressing content and style gaps with adaptive thresholds and bridging layers, and extending it to video for improved performance. Their approach demonstrated consistent cross-domain performance, unlike train-based baselines that specialize only in their trained domains.
We present a novel, training-free approach to scene change detection. Our method leverages tracking models, which inherently perform change detection between consecutive frames of video by identifying common objects and detecting new or missing objects. Specifically, our method takes advantage of the change detection effect of the tracking model by inputting reference and query images instead of consecutive frames. Furthermore, we focus on the content gap and style gap between two input images in change detection, and address both issues by proposing adaptive content threshold and style bridging layers, respectively. Finally, we extend our approach to video, leveraging rich temporal information to enhance the performance of scene change detection. We compare our approach and baseline through various experiments. While existing train-based baseline tend to specialize only in the trained domain, our method shows consistent performance across various domains, proving the competitiveness of our approach.