AI & Vision · Project case study

Facial expression recognition

Learning visual expression patterns with a CNN.

DisciplineAI & Vision
StatusCompleted
Year2026

Project overview

PythonPyTorchCNNFER-2013Computer vision
65.4%Reported FER-2013 validation accuracy
VISION PIPELINE / CONCEPT DIAGRAMIMAGELEARNED FEATURESCLASSES

A convolutional neural network trained on FER-2013 to classify facial expressions. The reported validation accuracy reached 65.4% after 75 epochs.

The engineering

The challenge

Classify facial expressions from images in the FER-2013 dataset using a convolutional neural network.

My contribution

Prepared image inputs, trained the CNN and evaluated its validation accuracy.

Approach & implementation

Applied image preprocessing, supervised CNN training and validation. Trained the model for 75 epochs using Python and PyTorch.

Results & lessons

What the project achieved

Reached a reported validation accuracy of 65.4% after 75 epochs.

Limitations & next steps

Validation accuracy describes performance on the validation split. Dataset expression labels do not establish a person’s actual emotional state.

Explore the engineering

Image input

A facial image is prepared for the model. Consistent input formatting helps the network learn from comparable data.

Interactive explanation of a classification pipeline. No images are analysed or uploaded by this diagram.

Let’s talk about this project.

I’m happy to walk through the design decisions, challenges and what I would improve next.

Contact Peter