Computer Vision Labs
Mar 2026
Finished
Three comprehensive computer vision lab assignments from the MUIARFID master's program. Exercise 1 — Gender Recognition: train a CNN with under 100K parameters achieving >98% accuracy. Supports small/large CNN and ResNet architectures. Exercise 2 — Car Model Identification: Bilinear CNN with pretrained backbones (VGG-16, ResNet-50) in three combinations (vg-res, vg-vg, res-res). Two-stage training with backbone freezing then full fine-tuning at reduced LR. Exercise 3 — Attention Visualization: explore ViT attention maps using raw attention rollout and interactive per-patch head selection. Built with PyTorch Lightning for clean training loops, automatic checkpointing, and early stopping. Includes 4 Jupyter notebooks for interactive experimentation and a CLI for batch training runs.
AI
HuggingFace
Jupyter
LaTeX
Matplot
NumPy
OpenCV
Plotly
Python
PyTorch
Sklearn
