# 🚀 CIFAR-100 ResNet-18 Hugging Face Space

This directory contains the Hugging Face Space for the CIFAR-100 ResNet-18 model.

## 📁 Files

- **`app.py`** - Main Gradio application
- **`requirements.txt`** - Python dependencies
- **`README.md`** - Space configuration and documentation
- **`test_model.py`** - Model testing script
- **`.gitignore`** - Git ignore patterns

## 🏆 Model Performance

- **77.45% Test Accuracy** (4.45% above 73% target)
- **11.22M Parameters** (ResNet-18 architecture)
- **100 CIFAR-100 Classes** supported

## 🔗 Links

- **Live Space:** https://huggingface.co/spaces/santhoshv6/ERA_V4_S8_Assignment
- **Model Repository:** https://github.com/santhoshv6/era_v4_s8_assignment
- **Model Download:** Available from GitHub releases

## 🚀 Local Testing

```bash
python test_model.py
python app.py
```

## 📦 Model Loading

The app automatically downloads the trained model from GitHub releases:
- URL: `https://github.com/santhoshv6/era_v4_s8_assignment/releases/download/v1.0/model_best.pth`
- Size: ~85.7 MB
- Format: PyTorch checkpoint with state_dict

## 🎯 Usage

1. Upload any image (will be resized to 32x32)
2. Get top 5 predictions with confidence scores
3. Model works best with CIFAR-100 style images

## 🔧 Technical Details

- **Framework:** PyTorch + Gradio
- **Preprocessing:** Resize to 32x32, normalize with CIFAR-100 stats
- **Architecture:** ResNet-18 with BasicBlocks
- **Classes:** All 100 CIFAR-100 fine-grained categories