Table 5.
FER solutions tested on a small number of human subjects and on the same database.
| Reference | Method | Database | Accuracy |
|---|---|---|---|
| [37] | DenseNet-161 | KDEF/4900 images JAFFE/213 images | 96.51% 98.78% |
| [38] | CNN | CK+/593 images JAFFE/213 images |
97.05% 98.63% |
| [88] | CNN | CASME II/247 images SMIC/164 images | 69.92% 54.84% |
| [61] | VGG16 | KDEF/4.900 images | 88% |
| [142] | CNN | JAFFE/213 images CK+/3150 images |
95.65% 99.36% |
| [102] | VGG19 | CK+/593 images JAFFE/213 images |
96.46% 91.27% |
| [72] | ResNet18 | CK+/593 video sequences AFEW 8.0/1.809 samples |
99.69% 51.18% |
| [89] | VGG-M, OC-NET | SMIC/164 images CASME II/145 images SAMM/132 images |
74.8% 90.8% 71.72% |
| [77] | GoogleLeNetv2 | CK+/593 sequences MMI/5130 images RaFD/67 images |
98.38% 99.59% 99.17% |
| [153] | ResNet101 | KDEF/4.900 images JAFFE/213 images RaFD/8.040 images |
94.59% 92.86% 98.88% |
| [154] | CNN | CK+/327 images JAFFE/213 images |
93.46% 94.75% |
| [155] | RNN | CK+/327 images | 95.4% |
| [65] | ResNet50 | CK+/593 images Oulu-CASIA/80 images AFEW/1809 images |
98.46% 87.31% 53.44% |