AnimalVision 🦁🐘🐊

AnimalVision is an animal image classifier that detects 64 different kinds of animals using a feature extractor based on EfficientNet-B2. You can see it in action here: AnimalVision on Hugging Face Spaces.

Model Overview

AnimalVision uses EfficientNet-B2 architecture with PyTorch's default pre-trained weights. It has been fine-tuned on a specialized dataset containing 64 classes of animal images from the Kaggle dataset by Anthony Therrien.

Key Features:

• 64 Animal Classes: The model can classify images into 64 different animal categories.
• EfficientNet-B2 Backbone: Lightweight and efficient architecture optimized for performance and accuracy.
• Trained on Kaggle Dataset: Fine-tuned on a dataset with over 64 animal species for improved real-world accuracy.

Training & Performance

The model was trained for 5 epochs. Below are the final training and testing performance results:

• Epoch: 5
• Training Loss: 0.7951
• Training Accuracy: 99.66%
• Testing Loss: 0.7759
• Testing Accuracy: 99.79%

How to Use on Hugging Face:

Simply access the web interface via Hugging Face Spaces and upload an image of an animal!

Dataset

The model was fine-tuned on the Kaggle dataset from Anthony Therrien. This dataset contains a variety of animal species with labeled images, making it ideal for multi-class classification tasks.

Acknowledgements

• EfficientNet: Pretrained weights provided by PyTorch.
• Kaggle: Animal images dataset available at Kaggle.