Image Classifier

This archive contains the code and prediction files for COMP30027 Project 2.

Files

• task1_animals.ipynb: code for Task 1 coarse-grained animal classification.
• task2_birds.ipynb: code for Task 2 fine-grained bird species classification.
• data/task1_data/resnet50_features.csv: precomputed ResNet-50 features for Task 1.
• data/task2_data/resnet50_features.csv: precomputed ResNet-50 features for Task 2.
• output/task1_resnet50_logreg_submission.csv: Kaggle-format prediction files generated by the models for Task 1.
• output/task2_resnet50_logreg_submission.csv: Kaggle-format prediction files generated by the models for Task 2.

Running the notebooks

Open and run the notebooks in order:

• task 1 code
• task 2 code

The notebooks load the provided feature files and the precomputed ResNet-50 feature files from the data/ folder for each task, train the models, evaluate them using validation/cross-validation, and generate Kaggle-format prediction CSV files.

Precomputed code blocks

Some computationally expensive blocks are commented out because they were run once during development and their outputs are saved in the archive.

• ResNet-50 feature extraction was run once and saved to:

  • data/task1_data/resnet50_features.csv
  • data/task2_data/resnet50_features.csv

• Hyperparameter tuning was run during model development. The selected hyperparameters are fixed in the final models so the notebooks can reproduce the final reported results directly.

Prediction files

Final Kaggle-format prediction files are located in the output/ folder.