An open API service indexing awesome lists of open source software.

https://github.com/supershivam07/vision-ai


https://github.com/supershivam07/vision-ai

cifar10 cnn confusion-matrix dataset evaluation fine-tuning inference jyputer-notebook keras mobilenetv2 model prediction preprocessing python tensorflow tensorflow2 training transfer-learning vision-ai weights

Last synced: 3 months ago
JSON representation

Awesome Lists containing this project

README

          

# Vision AI in 5 Days – CIFAR-10

This repo is a complete starter to build, evaluate, and demo an image-classification system in **Python + TensorFlow/Keras**. It follows the bootcamp flow: preprocess → train basic CNN → evaluate → transfer learning → fine-tune → demo → publish.

## Quickstart (Google Colab recommended)

1. **Upload this zip** to Colab and unzip:
```bash
!unzip vision_ai_cifar10_project.zip -d /content/
%cd /content/vision_ai_cifar10_project
```

2. **(Optional) Preprocess & cache data:**
```bash
!python src/preprocess.py
```

3. **Train a basic CNN** (saves model & plots):
```bash
!python src/train_cnn.py
```

4. **Evaluate saved model** (confusion matrix + report):
```bash
!python src/evaluate.py --model models/basic_cnn.h5 --prefix basic_cnn
```

5. **Transfer learning (MobileNetV2)**:
```bash
!python src/train_transfer.py
!python src/evaluate.py --model models/mobilenetv2_cifar10.h5 --prefix mobilenetv2
```

6. **Single-image inference**:
```bash
# Replace with your image path
!python src/infer.py --model models/mobilenetv2_cifar10.h5 --image sample.jpg
```

## Repo Layout
```
vision_ai_cifar10_project/
├─ data/ # cached dataset (created automatically)
├─ docs/ # slides, LinkedIn post, demo script templates
├─ models/ # trained .h5 models
├─ outputs/ # plots, confusion matrices, reports
└─ src/ # all Python code
```

## Notes
- Default dataset: **CIFAR-10** (built into Keras). You can swap to **Cats vs Dogs** via `tensorflow_datasets` with small changes.
- Training uses **GPU** if available. In Colab: Runtime → Change runtime type → GPU.
- For fast submission, you can reduce epochs in the training scripts.

Good luck! 🚀