Volume 10,Issue 9
Fine-Grained Visual Classification: From Part Discovery to Efficient Multimodal Reasoning
This survey examines FGVC, in which a model must distinguish visually similar categories that nevertheless exhibit substantial within-class variation. We organize recent progress around three decisions that recur across architectures: which evidence should be selected, how local and global evidence should be integrated, and how supervision should be obtained when detailed labels are scarce. We first define the task and examine the assumptions embedded in commonly used benchmarks. We then trace the transition from convolutional part discovery and second-order feature encoding toward token-based transformer models, focusing on design principles that generalize across implementations rather than presenting a chronological catalog. The review further compares localization, feature interaction, augmentation, and parameter-efficient adaptation, and discusses how evaluation protocols affect reported gains. We conclude by examining annotation bias, long-tailed distributions, domain shift, computational constraints, and the emerging role of vision-language models. The resulting synthesis is intended to support method selection and to identify research problems that extend beyond marginal improvements on saturated benchmarks.
[1] Fu J, Zheng H, Mei T, 2017, Look Closer to See Better: Recurrent Attention Convolutional Neural Network for Fine-Grained Image Recognition. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 4438–4446.
[2] Yang Z, Luo T, Wang D, et al., 2018, Learning to Navigate for Fine-Grained Classification. Proceedings of the European Conference on Computer Vision (ECCV), 420–435.
[3] Luo W, Yang X, Mo X, et al., 2019, Cross-X Learning for Fine-Grained Visual Categorization. Proceedings of the IEEE/CVF International Conference on Computer Vision, 8242–8251.
[4] Dosovitskiy A, Beyer L, Kolesnikov A, et al., 2020, An Image Is Worth 16x16 Words: Transformers for Image Recognition at Scale. arXiv preprint arXiv:2010.11929.
[5] He J, Chen J, Liu S, et al., 2022, TransFG: A Transformer Architecture for Fine-Grained Recognition. Proceedings of the AAAI Conference on Artificial Intelligence, 36(1): 852–860.
[6] Xu Q, Wang J, Jiang B, et al., 2023, Fine-Grained Visual Classification via Internal Ensemble Learning Transformer. IEEE Transactions on Multimedia, 25: 9015–9028.
[7] Chen C, Fan Q, Panda R, 2021, CrossViT: Cross-Attention Multi-Scale Vision Transformer for Image Classification. Proceedings of the IEEE/CVF International Conference on Computer Vision, 357–366.
[8] Sun H, He X, Xu J, et al., 2024, SIM-OFE: Structure Information Mining and Object-Aware Feature Enhancement for Fine-Grained Visual Categorization. IEEE Transactions on Image Processing.
[9] Huang Y, Hechen Z, Zhou M, et al., 2025, An Attention-Locating Algorithm for Eliminating Background Effects in Fine-Grained Visual Classification. IEEE Transactions on Circuits and Systems for Video Technology.
[10] Shi Y, Hong Q, Yan Y, et al., 2025, LDH-ViT: Fine-Grained Visual Classification Through Local Concealment and Feature Selection. Pattern Recognition, 161: 111224.
[11] Hinton G, Vinyals O, Dean J, 2015, Distilling the Knowledge in a Neural Network. arXiv preprint arXiv:1503.02531.
