Learning project for school.
Idea is to present your work in detail to other students in the course.
EfficientNet-B4 (pretrained)
High quality birds dataset from kaggle: https://www.kaggle.com/datasets/gpiosenka/100-bird-species/data
- 525 bird species -> reduced to 184 (Only bird names starting with A,B or C)
- 84635 training img -> reduced to 29274
- Each species has at least 130 training imgs
- ~80% Male and 20% female, this is a shortcoming in the dataset
- 2625 test & validation imgs (~3% each, very low IMO) -> reduced to 920
- 224x224x3 JPG
Data is prepared, we can go almost straight into modeling and evaluation.
Plan is to make a nice well explained Jupyter Notebook that can be presented live (except the training of course)
- Save the model
- Explain everything
- Visualizations
Note: Again, this is a learning project. Never coded PyTorch myself before this.
Materials I've used so far:
- https://pytorch.org/tutorials/beginner/basics/quickstart_tutorial.html (Low complexity)
- https://www.kaggle.com/code/ibrohimsharipov/birds-classification-pytorch-96-accuracy (Easy to follow)
- https://www.kaggle.com/code/harpdeci/classifying-birds-with-resnet (This one pretty advanced)
- ChatGPT4 (no copypasta code, just further explain some concepts for me)
- Google search