An efficient multi-scale channel attention network for person re-identification

Abstract

At present, occlusion and similar appearance pose serious challenges to the task of person re-identification. In this work, we propose an efficient multi-scale channel attention network (EMCA) to learn robust and more discriminative features to solve these problems. Specifically, we designed a novel cross-channel attention module (CCAM) in EMCA and placed it after different layers in the backbone. The CCAM includes local cross-channel interaction (LCI) and channel weight integration (CWI). LCI focuses on both the maximum pooling features and the average pooling features to generate channel weights through convolutional layers, respectively. CWI combines the two channel weights to generate richer and more discriminant channel weights. Experiments on four popular person Re-ID datasets (Market-1501, DukeMTMC-ReID, CUHK-03 (detected) and MSMT17) show that the performance of our EMCA is consistently significantly superior to the existing state-of-the-art methods.

Person Re-Identification Across Multiple Cameras: An Efficient Gated Channel-Spatial Attention Method

“…The SFTCA mechanism generates a 2D attention map based on channel adaptability and spatial location, allowing iReIDNet to focus on salient information in pedestrian images. Luo et al [22] proposed an efficient multi-scale channel attention network (EMCA) that aims to learn robust and discriminative features. The EMCA includes cross-channel attention module (CCAM) which incorporates local cross-channel interaction (LCI) and channel weight integration (CWI) to enhance feature learning.…”

“…Recently, Re-ID models [20][21][22][23]have been incorporating attention modules to extract distinctive features and address challenges such as occlusion and similar appearances.…”

“…Chen et al [21] proposed a feature pyramid network (APNet) that adopts the "Squeeze-and-Excitation" block (SE) [25] for channel-wise attention to give more weight to channels with a stronger response, and RGANet [26] for spatial attention to achieve the same effect spatially. Lou et al [22]proposed an efficient multi-scale channel attention network (EMCA) to learn robust and more discriminative features, using global max and average pooling to solve the occlusion problem. Chen et al [27] introduce the Occlusion-Aware Mask Network (OAMN) which includes an attention-guided mask module that relies on labeled occlusion data to generate diverse and accurately labeled occlusion for any dataset, making it more suitable for real-world scenarios compared to existing methods.…”