Comparative Performance Analysis of CNN, Vision Transformer, and Hybrid Model Architectures in Digital Image Classification
Kata Kunci:
image classification; CNN; Vision Transformer; hybrid model; deep learningAbstrak
Digital image classification is a fundamental computer vision task whose development is driven by three main architectural paradigms: the Convolutional Neural Network (CNN), the Vision Transformer (ViT), and hybrid models combining both. Selecting the right architecture is challenging because each has distinct strengths and limitations. This study aims to comparatively analyze the performance of the three paradigms based on a literature review of publications from 2022 to 2026. The method is a literature-based comparative study that synthesizes performance metrics reported across various studies, including accuracy, parameter efficiency, inference speed, and generalization capability. The analysis shows that CNNs excel in data efficiency and local feature extraction, making them suitable for resource-constrained devices; ViTs achieve superior accuracy on large-scale datasets through global attention but require large-scale pretraining; while hybrid models such as CoAtNet consistently achieve the highest accuracy (up to 88.6% Top-1 on ImageNet) by combining the strengths of both. This study concludes that no architecture is universally superior; architecture selection must consider dataset characteristics, computational budget, and deployment context
Referensi
1. Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., ... & Houlsby, N. (2021). An image is worth 16x16 words: Transformers for image recognition at scale. In International Conference on Learning Representations (ICLR).
2. Liu, Z., Mao, H., Wu, C. Y., Feichtenhofer, C., Darrell, T., & Xie, S. (2022). A ConvNet for the 2020s. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (pp. 11976–11986).
3. Dai, Z., Liu, H., Le, Q. V., & Tan, M. (2021). CoAtNet: Marrying convolution and attention for all data sizes. Advances in Neural Information Processing Systems, 34, 3965–3977.
4. Papastratis, I. (2023). Comparison of Convolutional Neural Networks and Vision Transformers (ViTs). Medium. Retrieved from https://medium.com/@iliaspapastratis
5. Aviral, A., et al. (2024). A comparative study of CNN, ResNet, and Vision Transformers for multi-classification of chest diseases. arXiv preprint arXiv:2406.00237.
6. Ashraf, S., et al. (2023). SynthEnsemble: A fusion of CNN, Vision Transformer, and hybrid models for multi-label chest X-ray classification. arXiv preprint arXiv:2311.07750.
7. Comparative study authors. (2025). Vision Transformers for kidney stone image classification: A comparative study with CNNs. arXiv preprint arXiv:2508.13461.
8. CoAtNeXt authors. (2025). CoAtNeXt: An attention-enhanced ConvNeXtV2-Transformer hybrid model for gastric tissue classification. arXiv preprint arXiv:2509.09242.
9. ConvTransGFusion authors. (2025). A hybrid learnable fusion of ConvNeXt and Swin Transformer for optimized image classification. IoT, 6(2), 30. https://doi.org/10.3390/iot6020030
10. Yuan, L., Chen, Y., Wang, T., Yu, W., Shi, Y., Jiang, Z., ... & Yan, S. (2022). Tokens-to-token ViT: Training Vision Transformers from scratch on ImageNet. arXiv preprint arXiv:2101.11986.
11. Liu, Z., Lin, Y., Cao, Y., Hu, H., Wei, Y., Zhang, Z., ... & Guo, B. (2022). Swin Transformer: Hierarchical Vision Transformer using shifted windows. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV).
12. Santopaolo, G. P. (2026). Benchmarking ResNet vs ViT on 50K images: A reality check. Retrieved from https://genmind.ch
13. EVCC authors. (2025). EVCC: Enhanced Vision Transformer-ConvNeXt-CoAtNet fusion for classification. arXiv preprint arXiv:2511.18691.
Diterbitkan
Terbitan
Bagian
Lisensi
Hak Cipta (c) 2026 Journal of Emerging Engineering Networks (JEEN)
Artikel ini berlisensiCreative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
