| # Real or Cake? | |
| Test it here : | |
| A terrible binary image classifier built on top of Andrej Karpathy's Micrograd and trained on a tiny handmade dataset of 100 images. | |
| The output is probably confidently wrong | |
| ## Why is it terrible? | |
| - dataset was frankly too small (100 images - 50 real objects (from shoes to plants) and 50 yummy cakes) | |
| - all images were resized to 8x8 px, leaving out a lot of visual information for the model | |
| - relies solely on micrograd's multi-layer perceptron (mlp) instead of convolutional neural networks or deep learning libarires like pytorch | |
| ## Accuracy | |
| Training: 70% | |
| Testing: 60% | |
| ## Run | |
| pip install -r requirements.txt | |
| python app.py | |
| weights.json is already included in the repository, so you don't need to retrain the model before launching the app |