| license: apache-2.0 | |
| tags: | |
| - image-classification | |
| - pytorch | |
| - cifar10 | |
| datasets: | |
| - uoft-cs/cifar10 | |
| # CIFAR-10 CNN Classifier | |
| A 3-block CNN trained from scratch on CIFAR-10 using PyTorch. | |
| ## Classes | |
| plane, car, bird, cat, deer, dog, frog, horse, ship, truck | |
| ## Training | |
| - Architecture: 3× Conv blocks (32→64→128 channels) + FC classifier | |
| - Optimizer: Adam + OneCycleLR scheduler | |
| - Augmentation: RandomCrop, RandomFlip, ColorJitter, RandomErasing | |
| - Validation accuracy: ~80%+ (after 64 epochs) | |
| ## Usage | |
| ```python | |
| from transformers import pipeline | |
| from PIL import Image | |
| classifier = pipeline( | |
| "image-classification", | |
| model="MarkivDunhar/cifar-cnn" | |
| ) | |
| result = classifier("your_image.jpg") | |
| print(result) | |
| # [{'label': 'dog', 'score': 0.91}] | |
| ``` | |