CIFAR-10 ResNet-18 Image Classifier
Description
ResNet-18 image classifier fine-tuned on CIFAR-10.
University laboratory: Przetwarzanie danych w chmurze.
Architecture
- Architecture: ResNet-18
- Framework: PyTorch
- Initial weights: ImageNet pretrained
- Training method: Transfer Learning
- Training samples: 8000
- Validation samples: 1000
- Test samples: 10000
- Training epochs: 2
Evaluation
- Test accuracy: 88.39%
- Macro F1-score: 0.8830
- Test loss: 0.3633
Classes
airplane, automobile, bird, cat, deer, dog, frog, horse, ship, truck
Files
- resnet18_best.pt — PyTorch checkpoint
- config.json — model and preprocessing configuration
- test_metrics.json — evaluation metrics
- per_class_metrics.csv — per-class evaluation
- confusion_matrix.png — confusion matrix
Limitations
This model was evaluated on CIFAR-10. Performance on other image distributions is unknown.
The checkpoint is intended for loading with PyTorch torchvision ResNet-18.