Developed by Raiyan Aaijaz & Heeya Shah
LateralNet explores biologically inspired lateral inhibition and excitation in convolutional networks to improve face recognition under occlusion. By modifying ResNet-18/50 with custom PyTorch layers, the project demonstrates improved robustness when faces are partially obscured (e.g., masks, sunglasses).
- Custom PyTorch layers simulating lateral inhibition/excitation
- Evaluation on the Real-World Occluded Faces (ROF) dataset
- Improved masked-face recognition accuracy by up to 17% over baseline
- Experiments comparing ResNet-18 vs ResNet-50 across occlusion types
notebooks/– Jupyter notebook containing training, evaluation, and visualizationsresults/– Accuracy tables and heatmapsdocs/– Project report
git clone https://github.com/raiyanjaz/lateralnet.git
cd lateralnet
pip install -r requirements.txt