Domain Adaptation via Incremental Confidence Samples into Classification
This repository contains the codes of the ICSC proposed in the International Journal of Intelligent Systems (IJIS).
Apart from this, I would like to introduce another repository that integrates some shallow domain adaptation methods. I hope this can help.
Code files (matlab implementation)
ICSC.m : the implementation of ICSC;
ACDA.m: the implementation of Adaptive Adjustment;
demo.m: experimental demonstration on Office (surf) dataset.
Results on datasets
There are some results conducted on different datasets can be directly cited.
1. Office+Caltech (Surf)
| SOURCE | TARGET | |
|---|---|---|
| C | A | 55.53 |
| C | W | 59.32 |
| C | D | 54.14 |
| A | C | 42.21 |
| A | W | 54.24 |
| A | D | 50.96 |
| W | C | 36.69 |
| W | A | 40.08 |
| W | D | 75.8 |
| D | C | 34.19 |
| D | A | 40.5 |
| D | W | 85.42 |
| AVERAGE | 52.42 |
2. Office+Caltech (Decaf)
| SOURCE | TARGET | |
|---|---|---|
| C | A | 93.32 |
| C | W | 93.22 |
| C | D | 92.36 |
| A | C | 88.78 |
| A | W | 89.15 |
| A | D | 96.18 |
| W | C | 83.7 |
| W | A | 88.41 |
| W | D | 99.36 |
| D | C | 82.9 |
| D | A | 89.67 |
| D | W | 98.98 |
| AVERAGE | 91.34 |
3. ImageNet +VOC2007
| SOURCE | TARGET | ICSC |
|---|---|---|
| I | V | 65.05 |
| V | I | 81.95 |
| AVERAGE | 73.5 |
4. OfficeHome (ResNet50)
| SOURCE | TARGET | |
|---|---|---|
| Ar | Cl | 51.7 |
| Ar | Pr | 71.3 |
| Ar | Re | 75.7 |
| Cl | Ar | 62 |
| Cl | Pr | 70.7 |
| Cl | Re | 70.7 |
| Pr | Ar | 62.4 |
| Pr | Cl | 50 |
| Pr | Re | 76 |
| Re | Ar | 68.2 |
| Re | Cl | 52.4 |
| Re | Pr | 79 |
| AVERAGE | 65.84 |
5. Office31 (ResNet50)
| SOURCE | TARGET | |
|---|---|---|
| A | D | 87.80 |
| A | W | 87.80 |
| D | A | 74.10 |
| D | W | 95.00 |
| W | A | 74.30 |
| W | D | 97.40 |
| AVERAGE | 86.07 |
6. ImageCLEF-DA (ResNet50)
| SOURCE | TARGET | |
|---|---|---|
| C | I | 93.00 |
| C | P | 79.86 |
| I | C | 95.00 |
| I | P | 80.71 |
| P | C | 94.17 |
| P | I | 93.00 |
| AVERAGE | 89.29 |
Citation
MLA: Teng, Shaohua, et al. "Domain adaptation via incremental confidence samples into classification." International Journal of Intelligent Systems 37.1 (2022): 365-385.
@article{teng2022domain,
title={Domain adaptation via incremental confidence samples into classification},
author={Teng, Shaohua and Zheng, Zefeng and Wu, Naiqi and Fei, Lunke and Zhang, Wei},
journal={International Journal of Intelligent Systems},
volume={37},
number={1},
pages={365--385},
year={2022},
publisher={Wiley Online Library}
}