DR-TANet: Dynamic Receptive Temporal Attention Network for Street Scene Change Detection

review

The paper introduces DR-TANet, It guarantees excellent performance with fewer parameters and less computational complexity .

Introduce

The most advanced relationship calculation method in change detection task is correlation calculation . Correlation calculation is carried out at a certain feature level or full feature level , Generated correlation diagram representation $t_0$ At a particular central location with $t_1$ Similarity between fixed square search areas around the same location . Although the correlation computer system can obtain better performance , But there is still much room for improvement .

The content of this paper mainly discusses the over fitting problem when detecting on the same data set and the difficult problem when detecting on different data sets .