Image classification is a basic and important direction in image processing. Since there is not only a single label value on an image, the current image classification can no longer meet people's needs, and multi-label image classification came into being. This paper proposes a multi-label image classification framework using Swin Transformer for feature extraction and a two-layer routing attention module for feature processing. Swin Transformer extracts multi-scale information through a hierarchical structure, and is superior to Vision Transformer in terms of multi-target and finer-grained image recognition. The dual-layer routing attention module enables more flexible computation allocation and content awareness. The dynamic attention mechanism adaptively adjusts the attention weight according to the characteristics of the input image, so that different positions or features can be given different levels of attention, and the intensity and range of attention can be flexibly controlled by adjusting the dynamic attention. The average precision of the model on the COCO dataset is 87.3, and the average precision on the VOC2007 dataset is 96.7, which improves the accuracy of multi-label image classification to a certain extent.
WANGJ, YANGY, MAOJ, et al. CNN-RNN: a unified framework for multi-label image classification[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 2285-2294.
[2]
CHENZ M, WEIX S, WANGP, et al. Multi-label image recognition with graph convolutional networks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019: 5172-5186.
[3]
KIPFT N, WELLINGM. Semi-supervised classification with graph convolutional networks[EB/OL].
[4]
LIQ, PENGX, QIAOY, et al. Learning label correlations for multi-label image recognition with graph networks[J]. Pattern Recognition Letters, 2020, 138: 378-384.
[5]
WUT, HUANGQ, LIUZ, et al. Distribution-balanced loss for multi-label classification in long-tailed datasets[C]//European Conference on Computer Vision, 2020: 162-178.
[6]
RIDNIKT, BEN-BARUCHE, ZAMIRN, et al. Asymmetric loss for multi-label classification[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021: 82-91.
[7]
ZHUK, WUJ. Residual attention: a simple but effective method for multi-label recognition[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021: 184-193.
[8]
WASWANIA, SHAZEERN, PARMARN, et al. Attention is all you need[C]// Conference and Workshop on Neural Information Processing Systems, 2017: 6000-6010.
[9]
ZHAOJ, YANK, ZHAOY, et al. Transformer-based dual relation graph for multi-label image recognition[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021: 163-172.
[10]
LIUS, ZHANGL, YANGX, et al. Query2Label: a simple transformer way to multi-label classification[EB/OL].
[11]
RIDNIKT, SHARIRG, BEN-COHENA, et al. ML-decoder: scalable and versatile classification head[C]//Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2023: 32-41.
[12]
LIUZ, LINY, CAOY, et al. Swin transformer: hierarchical vision transformer using shifted windows[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021: 9992-10002.
[13]
ZHUL, WANGX, KEZ, et al. BiFormer: vision transformer with bi-level routing attention[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023: 10323-10333.
[14]
DOSOVITSKIYA, BEYERL, KOLESNIKOVA, et al. An image is worth 16×16 words: transformers for image recognition at scale[EB/OL].