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

Abstract

Street scene change detection continues to capture researchers' interests in the computer vision community. It aims to identify the changed regions of the paired street-view images captured at different times. The state-of-the-art network based on the encoder-decoder architecture leverages the feature maps at the corresponding level between two channels to gain sufficient information of changes. Still, the efficiency of feature extraction, feature correlation calculation, even the whole network requires further improvement. This paper proposes the temporal attention and explores the impact of the dependencyscope size of temporal attention on the performance of change detection. In addition, based on the Temporal Attention Module (TAM), we introduce a more efficient and light-weight version -Dynamic Receptive Temporal Attention Module (DRTAM)and propose the Concurrent Horizontal and Vertical Attention (CHVA) to improve the accuracy of the network on specific challenging entities. On street scene datasets 'GSV', 'TSUNAMI' and 'VL-CMU-CD', our approach gains excellent performance, establishing new state-of-the-art scores without bells and whistles, while maintaining high efficiency applicable in autonomous vehicles.

Domain Invariant Siamese Attention Mask for Small Object Change Detection via Everyday Indoor Robot Navigation

“…Recently, several studies have reported the effectiveness of attention in change detection tasks. HPCFNet [28] represents attention as a correlation between feature maps, DR-TA Net [15] evaluates temporal attention by computing the similarity and dependency between a feature map pair, to realize attention-based change detection. CSCDNet [11] employs a correlation filter to compensate for the uncertainty in the nonlinear transformation between live and reference images.…”

“…A major advantage of our proposed approach, owing to its reliance on highlevel contextual attention information rather than low-level visual features, is its potential to operate effectively in test domains with unseen complex backgrounds. In this sense, our approach combines the advantages of two major research directions in the change detection community: pixel-wise differencing [11], [15] and context-based novelty detection [9], [16], by incorporating all available information into the attention mechanism.…”

“…Our contributions can be summarized as follows: (1) We explore a new approach, called domain-adaptive attention model, to image change detection for robotics applications, with an ability of unsupervised on-the-fly domain adaptation. (2) Instead of considering pixel-wise differencing [11], [15] and context-based novelty detection [9], [16], as two independent approaches, our framework combines the advantages of both approaches by incorporating all available information into the attention mechanism. (3) We present a practical system for image change detection using stateof-the-art techniques such as image registration [17], pixel warping [18], and Siamese ConvNet [11].…”

Optimized Deep Belief Neural Network for Semantic Change Detection in Multi-Temporal Image

“…So far, changes of the land surface are detected by utilizing the significant data source called multi-temporal remote sensing images in a wide geographical area, and the requirement for conventional field researches is progressively decreased. Here, the most important thing is change detection (CD) and it is described as the detecting operation of changes which is obtained among two or more images that are based on the image properties [5]. The image property's differences such as texture, shape, and value of pixel radiance are can be interconnected to the changes on the ground at various satellite remark times [3].…”