The task was to develop a CNN that could accurately identify bird species from images. We used the Kaggle environment for development. Our code was written in Python using using Keras/TensorFlow. The final version reached 71.30% accuracy on test data. For developmemt we used our PCs and the freely available Kaggle GPU resources. No pretrained models were used.
Our group worked on this university project in summer semester of 2022.
- View project on Kaggle (some elements hidden for readibilty)
- View project on GitHub (full project as python notebook)
- Data handlding with standard libraries (Pandas, Numpy)
- Image import and preprocessing with Keras/Tensorflow
- Model build with Keras/TensorFlow
- Some metrics for evaluation with Sklearn
More information in the protocol
- BIRDS 400 - SPECIES IMAGE CLASSIFICATION from Kaggle