Dino Classification Challenge

Team Project

Team Project (AI/ML Course)

Dino Classification Challenge

Team Project (AI/ML Course)

DinoDL is a deep learning project focused on classifying dinosaur images into 7 categories using TensorFlow and Keras. We explored multiple models including a custom CNN and transfer learning with pretrained architectures (EfficientNet, ResNet, MobileNet, etc). Applied preprocessing techniques like resizing, augmentation, and normalization. Evaluated performance using confusion matrices and visualized training curves. Submitted final predictions via Kaggle.

Key Learnings:

Designed and trained CNN models from scratch
Applied data augmentation techniques for generalization
Implemented transfer learning using EfficientNet, ResNet, DenseNet, MobileNet
Visualized model performance with training curves and confusion matrices
Used modular function-based approach for repeatable training
Submitted predictions to Kaggle for evaluation

Screenshots

Dino Classification Challenge screenshot 1

Dino Classification Challenge screenshot 2

Dino Classification Challenge screenshot 3

Dino Classification Challenge screenshot 4

Dino Classification Challenge screenshot 5

Dino Classification Challenge screenshot 6

Technologies Used:

TensorFlowKerasPythonPandasNumPyMatplotlib

My Contribution

I worked specifically on the transfer learning part and refined pre-trained models for our requirements. I led data preprocessing, implemented training and evaluation pipelines, and contributed to the comparison of pretrained models. I also visualized the results using confusion matrices and loss/accuracy graphs and created the final Kaggle submission file.