Scale Adaptive Network for Partial Person Re-identification: Counteracting Scale Variance

HongYu Chen (Northwestern Polytechnical University),* BingLiang Jiao (Northwestern Polytechnical University ), Liying Gao ( Northwestern Polytechnical University), Peng Wang (Northwestern Polytechnical University)
The 34th British Machine Vision Conference


Partial Person Re-identification (Partial ReID) is a challenging task which aims to match partially visible images with holistic images of the same pedestrian. One of the significant challenges of this task is scale misalignment between holistic and partial person images, which makes it difficult for models to adapt to the scale gaps of different images. Previous methods used pooling or convolutional layers with various sizes to extract features of different scales. However, it is essential to note that each person's image has specifically suitable feature extraction scales, and some scale features may be unnecessary or even detrimental. Based on this finding, an adaptive feature extraction paradigm could be more suitable for Partial ReID. To this end, we propose a novel Scale Adaptive Network (SANet) to dynamically extract scale-adaptive features to counteract scale variance. Specifically, we introduce an Adaptive Feature Enhancement module (AFE) to adaptively extract multi-scale features and address scale misalignment. Furthermore, since a partial image only contains a portion of body parts in holistic images, the body parts exclusive to holistic images could introduce noise for image matching. Thus, we utilize a segmentation head to indicate the available human parts in each image and use the common visible body parts for feature comparisons between images. Extensive experiments demonstrate the effectiveness of our SANet network, which achieves state-of-the-art performance on partial and holistic person ReID benchmarks.



author    = {HongYu Chen and BingLiang Jiao and Liying Gao and Peng Wang},
title     = {Scale Adaptive Network for Partial Person Re-identification: Counteracting Scale Variance},
booktitle = {34th British Machine Vision Conference 2023, {BMVC} 2023, Aberdeen, UK, November 20-24, 2023},
publisher = {BMVA},
year      = {2023},
url       = {}

Copyright © 2023 The British Machine Vision Association and Society for Pattern Recognition
The British Machine Vision Conference is organised by The British Machine Vision Association and Society for Pattern Recognition. The Association is a Company limited by guarantee, No.2543446, and a non-profit-making body, registered in England and Wales as Charity No.1002307 (Registered Office: Dept. of Computer Science, Durham University, South Road, Durham, DH1 3LE, UK).

Imprint | Data Protection