Enhancing Semi-Supervised Semantic Segmentation via Image Search and Advanced Pooling Strategies ShuaiKangYang, Yu Gu, Lidong Yang, et al. STI: Enhancing Semisupervised Semantic Segmentation via Image Search and Advanced Pooling Strategies. Concurrency and Computation: Practice and Experience, 2026, 38(8): e70698. https://doi.org/10.1002/cpe.70698.
ResNet50 and DeepLabv3+
| Method | mIoU-1/16(%) | mIoU-1/8(%) | mIoU-1/4(%) |
|---|---|---|---|
| SupOnly | 62.99 | 65.08 | 68.85 |
| ESC | - | 70.2 | 72.6 |
| DCC | 70.1 | 72.4 | 74.0 |
| MT | 66.77 | 70.78 | 73.22 |
| CCT | 65.22 | 70.87 | 73.43 |
| GCT | 64.05 | 70.47 | 73.45 |
| CPS | 68.21 | 73.20 | 74.24 |
| CTT | - | 73.66 | 75.07 |
| ELN | - | 70.34 | 73.52 |
| USCS | 72.30 | 74.88 | 76.15 |
| RRN | 73.38 | 74.91 | 76.80 |
| PGCL | - | 75.20 | 76.00 |
| CPCL | 71.66 | 73.74 | 75.35 |
| FPL | 72.52 | 73.74 | 75.35 |
| RWMS | 72.20 | 75.03 | 76.63 |
| ST++ | 72.6 | 74.4 | 75.4 |
| STI(ours) | 73.75 | 75.98 | 76.14 |
ResNet101 and DeepLabv3+
| Method | mIoU-1/16(%) | mIoU-1/8(%) | mIoU-1/4(%) |
|---|---|---|---|
| SupOnly | 64.97 | 67.57 | 70.45 |
| AdvSeg | 68.2 | 69.5 | - |
| MT | 69.8 | 71.5 | 73.0 |
| S4GAN | 69.1 | 72.4 | 74.5 |
| GCT | 67.2 | 72.5 | 75.1 |
| CCT | 70.8 | 72.2 | 75.1 |
| PseudoSeg | - | 73.2 | - |
| DCC | 72.4 | 74.6 | 76.3 |
| PC2Seg | - | 74.1 | - |
| CPS | 69.8 | 74.3 | 74.6 |
| AEL | 74.5 | 75.6 | 77.5 |
| CutMix | 67.98 | 69.15 | 73.66 |
| U2PL | 74.9 | 76.5 | 78.5 |
| ST | 72.9 | 75.7 | 76.4 |
| ST++ | 74.5 | 76.3 | 76.6 |
| STI(ours) | 75.51 | 76.90 | 76.98 |
To run our code, you may need one GeForce RTX 3090(24G memory).
You can download cls_txt file, spilts file and pretrained file from CISC-R
python train.py
python eval.pyTo ensure the code can run, we provide versions of some libraries.
- apex-0.1
- python-3.8.13
- numpy-1.23.2
- torch-1.8.1
- pandas-1.5.3
- opencv-python-4.8.1
If there are any missing citations, please contact us. It is an unintentional omission, and we will add the citations accordingly.
This code is based on the implementation of ST++, CISC-R, Cutout and SoftPool.
If there are any missing citations, please contact us. It is an unintentional omission, and we will add the citations accordingly.
- Yang L, Zhuo W, Qi L, Shi Y, Gao Y.: St++: Make self-training work better for semi-supervised semantic segmentation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 4268-4277 (2022).
- Wu L, Fang L, He X, He M, Ma J, Zhong Z.: Querying Labeled for Unlabeled: Cross-Image Semantic Consistency Guided Semi-Supervised Semantic Segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(7):8827-8844 (2023).
- DeVries, Terrance. "Improved Regularization of Convolutional Neural Networks with Cutout." arxiv preprint arxiv:1708.04552 (2017).
- Stergiou, A., Poppe, R., & Kalliatakis, G. Refining activation downsampling with SoftPool. In Proceedings of the IEEE/CVF international conference on computer vision, pp. 10357-10366 (2021).