a computer vision project developed for classifying dog and cat breeds using the Oxford-IIIT Pet Dataset. 🐶🐱
Dataset: 37 dog/cat breeds with 200 images each.
Augmentation: Increased robustness through data augmentation, creating a richer dataset for model training.🔄
Initial Attempts: Custom model architectures; trialed VGG-16 and VGG-19.
Final Choice: ResNet50-V2, which significantly improved accuracy.🚀
Architecture: ResNet50-V2 (pre-trained, partially frozen), followed by fully connected layers, dropout, and a final output layer.
Validation Accuracy: Achieved a high accuracy (specific percentage detailed in the project).✅
Optimization: Used ADAM optimizer for better performance.⚙️
Learning Rate Scheduler: Implemented to optimize training process.📉
Tip
Validation Accuracy: 87%
Validation Loss: 41%
| Keeshond | Birman |
|---|---|
 |
 |
Confusion Matrix: Indicates good performance, with some confusion between similar breeds.🧩
Sample Outputs: Showcases the model's breed classification capabilities.👀
Data Distribution: Increasing the representation of confused breeds could enhance accuracy. However, dataset limitations restrict this approach.📚
Further Tuning: Optimizing learning rate and further experimenting with model architecture and training strategies.🔧
- Clone the repo.
- Install dependencies (list dependencies here).
- Run the model (provide command or script).
Feel free to contribute! Open an issue or submit PRs.
MIT License