EUVIMEDEUVIMED entdecken
Crossref

3 × 3 Kernel Is All You Need for Vision

Shenqi Lai, Mengjian Li, Haifeng Liu, Xueming Qian, Deng Cai, Yaxiong Wang

IEEE Transactions on Circuits and Systems for Video Technology · 2026-08 · 36 (8) · 11363-11375

DOI: 10.1109/tcsvt.2026.3686189 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Circuits and Systems for Video Technology, 2026-08. Verfasst von Shenqi Lai und 5 weiteren Autor:innen. Die Arbeit nennt 79 Literaturangaben. Gefördert von: National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Most modern Convolutional Neural Networks (CNNs) employ a multi-branch structure with various-sized convolutions to capture long- and short-range dependencies. However, these CNNs use large kernel convolutions (e.g., astonishingly 101 kernels) and specialized techniques (e.g., re-parameterization and sparsity), increasing complexity in both training and inference stages. This paper focuses on designing an efficient CNN based on pure $3\times 3$ convolutions without introducing complex operations and techniques. Specifically, we propose a Spatial Pyramid (SP) block, which consists of the Multi-branch Residual (MbR) module and the Gated-branch Residual (GbR) module. The MbR introduces multiscale pooling as the key component, thus capturing long-range visual cues through large down-sampling rates and shorter-range dependencies through low down-sampling rates while maintaining low computational complexity. Besides, the GbR uses one $3\times 3$ convolution to refine dependencies along spatial and channel dimensions. Based on the SP block, we construct the Spatial Pyramid CNN (SPCNN), a model composed exclusively of Point-Wise Convolution and $3\times 3$ Depth-Wise Convolution. Under comparable computational complexity, SPCNN significantly outperforms the state-of-the-art CNN PeLK (83.6% vs 82.6%) with only $3\times 3$ kernels (compared to $101\times 101$ kernels in PeLK). Besides, our SPCNN demonstrates comparability with state-of-the-art backbones in lightweight models, object detection, instance segmentation, and semantic segmentation. Moreover, evaluations of four image retrieval benchmarks also demonstrate the effectiveness. All codes are released at https://github.com/xiaolai-sqlai/SPCNN

Förderung

  • National Natural Science Foundation of China

Zitieren

APA

Lai, S., Li, M., Liu, H., Qian, X., Cai, D., & Wang, Y. (2026). 3 × 3 Kernel Is All You Need for Vision. IEEE Transactions on Circuits and Systems for Video Technology, 36(8), 11363-11375. https://doi.org/10.1109/tcsvt.2026.3686189

Vancouver

Lai S, Li M, Liu H, Qian X, Cai D, Wang Y. 3 × 3 Kernel Is All You Need for Vision. IEEE Transactions on Circuits and Systems for Video Technology. 2026;36(8):11363-11375. doi:10.1109/tcsvt.2026.3686189

BibTeX
@article{Lai2026,
  author = {Lai, Shenqi and Li, Mengjian and Liu, Haifeng and Qian, Xueming and Cai, Deng and Wang, Yaxiong},
  title = {3 × 3 Kernel Is All You Need for Vision},
  journal = {IEEE Transactions on Circuits and Systems for Video Technology},
  year = {2026},
  volume = {36},
  number = {8},
  pages = {11363--11375},
  doi = {10.1109/tcsvt.2026.3686189}
}
Kostenlos starten →