// project
CompletedSolo Developer
Marine Life Classification with Deep Learning
PyTorchViTEfficientNetComputer Vision
CS5242 Course Project — repo to be added
Overview
- Built an end-to-end classification pipeline on 20k+ real-world underwater species images, including data cleaning, augmentation, and handling of class imbalance.
Results
- Fine-tuned a Transformer-based ViT to 97.4% test accuracy and a CNN-based EfficientNet-B0 to 87% with far fewer parameters; analyzed the accuracy–efficiency trade-off between the two.