This project utilizes transfer learning on the VGG16 model for image classification in the CK Plus dataset. Transfer learning involves pre-training a neural network on a large and complex dataset, such as ImageNet, and using this pre-trained network for classifying images in a different dataset.
The VGG16 model is a convolutional neural network pre-trained on ImageNet, with a complex architecture capable of extracting deep features from the images to be classified. Through transfer learning, we can leverage the knowledge accumulated by VGG16 during pre-training on ImageNet and apply it to classify images in the CK Plus dataset.
The CK Plus dataset contains a collection of facial images of multiple subjects, with different facial expressions. Each image is associated with a label indicating the type of facial expression present in the image. The goal of the project is to train the VGG16 model to correctly classify facial expressions in images.
To use transfer learning with the VGG16 model, follow these steps:
- Load the pre-trained VGG16.
- Freeze the base layers of the VGG16 model to retain the pre-trained knowledge.
- Add a custom classification layer to the model, tailored to the CK Plus dataset.
- Train the model using the CK Plus dataset, fine-tuning the parameters of the added classification layer.
- Evaluate the performance of the model on the test dataset and analyze the results obtained.
| Epochs | Train accuracy | Test accuracy | Train time | Test time |
|---|---|---|---|---|
| 10 | 86.88% | 83.05% | 143.06 s | 5.33 s |
| 20 | 95.19% | 89.49% | 203.06 s | 5.33 s |
| 30 | 96.94% | 93.56% | 256.66 s | 5.33 s |
| 40 | 99.42% | 95.93% | 383.09 s | 3.81 s |
| 50 | 99.71% | 98.31% | 443.16 s | 5.34 s |