SEResNet10 is an image classification model based on ResNet with Squeeze-and-Excitation Blocks and SiLU.
Seems to work fine.
There are 10 layers/blocks in total.
Model in /weights is trained upon this dataset:
Aerial Landscape Images https://www.kaggle.com/datasets/ankit1743/skyview-an-aerial-landscape-dataset
Comparation with YOLO11-cls:
| Model | Image size (Pixels) |
Accuracy | Params | Flops |
|---|---|---|---|---|
| SEResNet10 | 256x256 | ~ 97.8% (@epoch=150) |
4.96M | 0.35G |
| EfficientNet | 256x256 | ~ 95.5% (@epoch=68) |
7.16M | 0.54G |
| YOLO11n-cls | 224x224 | ~ 89.4% | 1.6M | 0.5G |
| YOLO11s-cls | 224x224 | ~ 92.7% | 5.5M | 1.6G |
$ pip install -r requirements.txtPlease refer to the follow link:
$ python3 train.py [-a | --augmentatioin] [-u | --unbalanced-dataset] [-h | --help][-a | --augmentation]: enable data augmentation
[-u | --unbalanced-dataset]: enable unbalanced dataset
[-h | --help]: print help menu
A running outcome will be generated under ./run/trainXX
$ python3 test.pyA running outcome will be generated under ./run/testXX
$ python3 detect.py <image_path> <weights_path>Weights file can be found at the repo's release.
Input → Linear → SiLU → Linear → Sigmoid → Output
[1] Elfwing, S., Uchibe, E. & Doya, K. 2018, 'Sigmoid-weighted linear units for neural network function approximation in reinforcement learning', Neurocomputing, vol. 298, pp. 166-174, doi: 10.1016/j.neucom.2018.01.063.
[2] Helmholtz, H. 1867, Handbuch der physiologischen Optik, Voss.
[3] He, K., Zhang, X., Ren, S. and Sun, J. (2015) 'Deep Residual Learning for Image Recognition'. arXiv preprint arXiv:1512.03385. Available at: https://arxiv.org/abs/1512.03385 [Accessed 27 Mar. 2025].
[4] Hu, J., Shen, L., Albanie, S., Sun, G. & Wu, E. 2018, 'Squeeze-and-Excitation Networks', IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 42, no. 8, pp. 2011-2023, doi: 10.1109/TPAMI.2019.2913372.
[5] LeCun, Y., Bottou, L., Bengio, Y. & Haffner, P. 1998, 'Gradient-based learning applied to document recognition', Proceedings of the IEEE, vol. 86, no. 11, pp. 2278-2324, doi: 10.1109/5.726791.
[6] Liu, Y., Cheng, M. and Lapata, M. (2017) Learning Structured Text Representations. arXiv preprint arXiv:1709.01507v4. Available at: https://arxiv.org/abs/1709.01507v4 [Accessed 27 Mar. 2025].
[7] Marr, D. 1982, Vision: A computational investigation into the human representation and processing of visual information, W.H. Freeman and Company.
[8] Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N., Kaiser, L. and Polosukhin, I. (2017) 'Attention is All You Need'. arXiv preprint arXiv:1706.03762. Available at: https://arxiv.org/abs/1706.03762 [Accessed 27 Mar. 2025].
[9] Zhang, D., Liu, H., Yang, Y., Wang, H., Liang, Y., Ye, Z., Qin, Z., Huang, T.S. and Zhuang, Y. (2018) 'Neural machine translation with deep attention'. Neural Networks, 100, pp. 24-35. doi: 10.1016/j.neunet.2017.12.010.