chicken-breeds-recognizer

An image classification model from data collection (scraping chicken types), cleaning, model training, deployment and API integration.
The model can classify 20 different types of chicken breeds
The types are following:

1. Austra White
2. Black Sex Link
3. Blue Laced Red Wyandotte
4. Bresse
5. Cherry Egger
6. Cochin
7. Cornish Cross
8. Cream Legbar
9. Easter Egger
10. Frizzle
11. Iowa Blue
12. Jersey Giant
13. Nankin
14. New Hampshire
15. Orpingtons
16. Polish
17. Shamo
18. Silkie
19. Silver Laced Wyandotte
20. Turken (Naked Neck)

Dataset preparations

Data Scraping: Collected the names of the chicken breeds through web scraping from this link.
Data Collection: Downloaded from DuckDuckgo using term name. Dataloader: Used fastai DataBlock API to set up the Dataloader. Data Augmentation: fastai provides default data augmentation which operates in GPU.
Details can be found in
.ipynb format notebooks/chicken_breed_recognizer_data_collections.ipynb
.py format src/chicken_breed_recognizer_data_collections.py

Training and Data Cleaning

Training: Fine-tuned a resnet34 model 3 times, for (10+10+5) epochs and got up to ~94% accuracy.
Here's the final Confusion Matrix.

Data Cleaning: This part was the most tedious and took the highest time. Since the data was collected from DuckDuckgo images, there were many noises, and also misguided images. Cleaned and updated data using fastai ImageClassifierCleaner. A visual tool for cleaning data. I cleaned the data each time after training/finetuning, except for the last time which was the final iteration of the model.

Deployment

Deployed the model to HuggingFace Spaces Gradio App. The implementation can be found in app folder or here

API integration with github pages

The deployed model API is intgegrated here in github pages website. Implementation can be found in docs folder.