Fruit Image Classification: A Model Comparison Study

Classification of fruit images using techniques such as the transfer learning, pruning, model trimming and custom model development. An attempt to adapt the model to multi-label classification with real-life examples.

Authors

Konrad Pawlik - [email protected]
Jan Fiszer - [email protected]

Conclusion

A thorough report detailing the project, the models selected and the decisions taken can be found in AML-Fruits360-Report.pdf

Prerequisite

• Python3
• numpy
• matplotlib
• pandas
• tensorflow
• tensorflow_model_optimization

Source code

1. exploratory_data_analysis.ipynb
2. transfer_learning.ipynb
3. custom_model.ipynb
4. multi_label_and_pruning.ipynb

Data

• data
  • fruits-360_dataset
    • fruits-360
      • Training
      • Test
      • test-multiple_fruits

The original dataset contains images of the fruit in the native resolution, but these have been omitted due to their scarcity.
Data can be downloaded from https://www.kaggle.com/datasets/moltean/fruits?datasetId=5857&sortBy=voteCount