Results
KinFaceW-I and KinFaceW-II datasets provide the following three settings to evaluate the performance of different algorithms:
- Unsupervised setting: No labeled kin relation information is used.
- Image-restricted setting: Only the given kin relation information is used in the training splits.
- Image-unrestricted setting: The identity information of the person is available to potentially form additional negative pairs in the training splits.
Finally, the mean accuracy and ROC curves are used for evaluation.
KinFaceW-I
Unsupervised Results
Table 1: The mean accuracy (%) under Unsupervised setting on the KinFace-I dataset.
| Methods | F-S | F-D | M-S | M-D | Mean |
| baseline: LBP [1] | 76.92 | 69.05 | 64.24 | 66.59 | 69.20 |
| baseline: HOG [1] | 79.53 | 70.93 | 67.66 | 72.84 | 72.74 |
Figure 1: ROC curves over 5 folds under Unsupervised setting on the F-S subset of KinFace-I.
Figure 2: ROC curves over 5 folds under Unsupervised setting on the F-D subset of KinFace-I.
Figure 3: ROC curves over 5 folds under Unsupervised setting on the M-S subset of KinFace-I.
Figure 4: ROC curves over 5 folds under Unsupervised setting on the M-D subset of KinFace-I.
Image-Restricted Results
Table 2: The mean accuracy (%) under Image-Restricted setting on the KinFace-I dataset.
| Methods | F-S | F-D | M-S | M-D | Mean |
| SILD (LBP) [2] | 78.22 | 69.40 | 66.81 | 70.10 | 71.13 |
| SILD (HOG) [2] | 80.46 | 72.39 | 69.82 | 77.10 | 74.94 |
| Polito [4] | 85.30 | 85.80 | 87.50 | 86.70 | 86.30 |
| LIRIS [4] | 83.04 | 80.63 | 82.30 | 84.98 | 82.74 |
| ULPGC [4] | 71.23 | 70.85 | 58.52 | 80.89 | 70.01 |
| NUAA [4] | 86.25 | 80.64 | 81.03 | 83.93 | 82.96 |
| BIU [4] | 86.90 | 76.48 | 73.89 | 79.75 | 79.25 |
Figure 5: ROC curves over 5 folds under Image-Restricted setting on the F-S subset of KinFace-I.
Figure 6: ROC curves over 5 folds under Image-Restricted setting on the F-D subset of KinFace-I.
Figure 7: ROC curves over 5 folds under Image-Restricted setting on the M-S subset of KinFace-I.
Figure 8: ROC curves over 5 folds under Image-Restricted setting on the M-D subset of KinFace-I.
Image-Unrestricted Results
Table 3: The mean accuracy (%) under Image-Unrestricted setting on the KinFace-I dataset.
| Methods | F-S | F-D | M-S | M-D | Mean |
| NRML (LBP) [3] | 81.43 | 69.76 | 67.23 | 72.87 | 72.82 |
| NRML (HOG) [3] | 83.68 | 74.64 | 71.56 | 79.96 | 77.46 |
| BIU (LBP) [4] | 85.51 | 76.54 | 69.93 | 74.36 | 76.59 |
| BIU (HOG) [4] | 86.90 | 76.48 | 70.62 | 79.75 | 78.44 |
Figure 9: ROC curves over 5 folds under Image-Unrestricted setting on the F-S subset of KinFace-I.
Figure 10: ROC curves over 5 folds under Image-Unrestricted setting on the F-D subset of KinFace-I.
Figure 11: ROC curves over 5 folds under Image-Unrestricted setting on the M-S subset of KinFace-I.
Figure 12: ROC curves over 5 folds under Image-Unrestricted setting on the M-D subset of KinFace-I.
KinFaceW-II
Unsupervised Results
Table 4: The mean accuracy (%) under Unsupervised setting on the KinFace-II dataset.
| Methods | F-S | F-D | M-S | M-D | Mean |
| baseline: LBP [1] | 75.40 | 66.60 | 70.60 | 66.00 | 69.65 |
| baseline: HOG [1] | 74.20 | 66.60 | 70.60 | 67.00 | 69.60 |
Figure 13: ROC curves over 5 folds under Unsupervised setting on the F-S subset of KinFace-II.
Figure 14: ROC curves over 5 folds under Unsupervised setting on the F-D subset of KinFace-II.
Figure 15: ROC curves over 5 folds under Unsupervised setting on the M-S subset of KinFace-II.
Figure 16: ROC curves over 5 folds under Unsupervised setting on the M-D subset of KinFace-II.
Image-Restricted Results
Table 5: The mean accuracy (%) under Image-Restricted setting on the KinFace-II dataset.
| Methods | F-S | F-D | M-S | M-D | Mean |
| SILD (LBP) [2] | 78.20 | 70.00 | 71.20 | 67.80 | 71.80 |
| SILD (HOG) [2] | 79.60 | 71.60 | 73.20 | 69.60 | 73.50 |
| Polito [4] | 84.00 | 82.20 | 84.80 | 81.20 | 83.10 |
| LIRIS [4] | 89.40 | 83.60 | 86.20 | 85.00 | 86.05 |
| ULPGC [4] | 85.40 | 75.80 | 75.60 | 81.60 | 80.00 |
| NUAA [4] | 84.40 | 81.60 | 82.80 | 81.60 | 82.50 |
| BIU [4] | 87.51 | 80.82 | 79.78 | 75.63 | 80.94 |
Figure 17: ROC curves over 5 folds under Image-Restricted setting on the F-S subset of KinFace-II.
Figure 18: ROC curves over 5 folds under Image-Restricted setting on the F-D subset of KinFace-II.
Figure 19: ROC curves over 5 folds under Image-Restricted setting on the M-S subset of KinFace-II.
Figure 20: ROC curves over 5 folds under Image-Restricted setting on the M-D subset of KinFace-II.
Image-Unrestricted Results
Table 6: The mean accuracy (%) under Image-Unrestricted setting on the KinFace-II dataset.
| Methods | F-S | F-D | M-S | M-D | Mean |
| NRML (LBP) [3] | 79.20 | 71.60 | 72.20 | 68.40 | 72.85 |
| NRML (HOG) [3] | 80.80 | 72.80 | 74.80 | 70.40 | 74.70 |
| BIU (LBP) [4] | 84.24 | 79.45 | 75.98 | 77.04 | 79.18 |
| BIU (HOG) [4] | 87.51 | 80.82 | 79.78 | 75.63 | 80.94 |
Figure 21: ROC curves over 5 folds under Image-Unrestricted setting on the F-S subset of KinFace-II.
Figure 22: ROC curves over 5 folds under Image-Unrestricted setting on the F-D subset of KinFace-II.
Figure 23: ROC curves over 5 folds under Image-Unrestricted setting on the M-S subset of KinFace-II.
Figure 24: ROC curves over 5 folds under Image-Unrestricted setting on the M-D subset of KinFace-II.
ROC Curves
Existing ROC files can be downloaded here:
Unsupervised
- baseline: LBP, unsup_baseline_LBP_KinFaceW-I, unsup_baseline_LBP_KinFaceW-II.
- baseline: HOG, unsup_baseline_HOG_KinFaceW-I, unsup_baseline_HOG_KinFaceW-II.
Image-Restricted
Image-Unrestricted
Methods
- Given a face pair, the cosine distance is used to predict whether there is a kin relation or not.
- Meina Kan, Shiguang Shan, Dong Xu, Xilin Chen.
Side-Information based Linear Discriminant Analysis for Face Recognition.
British Machine Vision Conference (BMVC), 2011.
- Jiwen Lu, Xiuzhuang Zhou, Yap-Peng Tan, Yuanyuan Shang, Jie Zhou.
Neighborhood Repulsed Metric Learning for Kinship Verification.
IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), 2014.
- Jiwen Lu, Junlin Hu, Venice Erin Liong, Xiuzhuang Zhou, Andrea Bottino, Ihtesham Ul Islam, Tiago Figuieiredo Vieira, Xiaoqian Qin, Xiaoyang Tan, Songcan Chen, Shahar Mahpod, Yosi Keller, Lilei Zheng, Khalid Idrissi, Christophe Garcia, Stefan Duffner, Atilla Baskurt, Modesto Castrillon-Santana, Javier Lorenzo-Navarro.
The FG 2015 Kinship Verification in the Wild Evaluation.
IEEE International Conference on Automatic Face and Gesture Recognition (FG), 2015.
[pdf]
† The ROC curves are provided upon request.