This project focuses on detecting emotions from images using various pre-trained deep learning models. The models are trained on raw data and evaluated based on accuracy, precision, and recall.
The following models were used for training, and their performance metrics are recorded below:
| Model | Trainer | Epochs | Val Accuracy | Val Precision | Val Recall | Test Accuracy | Test Precision | Test Recall |
|---|---|---|---|---|---|---|---|---|
| Xception | Monya | 10 | 44.98% | - | - | - | - | - |
| 20 | 48.44% | - | - | - | - | - | ||
| MobileNet_V2 | Advika | 10 | 64.04% | 0.64 | 0.64 | - | - | - |
| 15 | 73.20% | 0.74 | 0.73 | - | - | - | ||
| 20 | 68.84% | 0.69 | 0.69 | 62.57% | 0.63 | 0.63 | ||
| ResNet18 | Advika | 10 | 62.07% | 0.62 | 0.62 | - | - | - |
| 15 | 61.79% | 0.62 | 0.62 | - | - | - | ||
| 20 | 61.67% | 0.61 | 0.62 | 61.44% | 0.60 | 0.61 | ||
| SqueezeNet1_1 | Advika | 10 | 56.29% | 0.57 | 0.56 | - | - | - |
| 15 | 58.46% | 0.58 | 0.58 | - | - | - | ||
| 20 | 58.76% | 0.61 | 0.59 | 57.70% | 0.60 | 0.58 | ||
| VGG16 | Ayushi | 15 | 32.63% | 0.17 | 0.23 | 33.63% | 0.18 | 0.22 |
| VGG19 | Ayushi | 15 | 29.79% | 0.19 | 0.24 | 30.44% | 0.18 | 0.23 |
| InceptionV3 | Manasvi | 10 | 37.33% | - | - | 40.00% | - | - |
| EfficientNetV2-B0 | Ayushi | 10 | 62.24% | 0.63 | 0.62 | - | - | - |
| 15 | 62.00% | 0.62 | 0.62 | - | - | - | ||
| 20 | 62.97% | 0.63 | 0.63 | 62.54% | 0.63 | 0.63 | ||
| ConvNeXt | Ayushi | 10 | 58.48% | 0.58 | 0.58 | - | - | - |
| 15 | 59.11% | 0.60 | 0.59 | - | - | - | ||
| 20 | 57.99% | 0.59 | 0.58 | 58.69% | 0.59 | 0.59 |