| --- |
| language: en |
| license: mit |
| tags: |
| - image-classification |
| - computer-vision |
| - pytorch |
| - cifar10 |
| datasets: |
| - cifar10 |
| metrics: |
| - accuracy |
| pipeline_tag: image-classification |
| --- |
| |
| # CIFAR-10 CNN Model |
|
|
| This is a convolutional neural network trained on the CIFAR-10 dataset, achieving 92.59% test accuracy after 100 epochs. |
|
|
| ## Model Details |
| - **Architecture**: 9 convolutional layers with batch normalization, max pooling, and dropout, followed by 3 fully connected layers. |
| - **Dataset**: CIFAR-10 (10 classes: airplane, automobile, bird, cat, deer, dog, frog, horse, ship, truck). |
| - **Training**: 100 epochs, SGD optimizer, CrossEntropyLoss, learning rate scheduling. |
| - **Accuracy**: 92.59% on the CIFAR-10 test set. |
|
|
| ## Usage |
| Load the model using: |
|
|
| ```python |
| from huggingface_hub import from_pretrained_pytorch |
| model = from_pretrained_pytorch('chandu1617/CIFAR10-CNN_Model') |
| ``` |
|
|
| ## Interactive Demo |
| Try the model in an interactive Gradio UI at [chandu1617/cifar10-cnn-demo](https://huggingface.co/spaces/chandu1617/cifar10-cnn-demo). |
|
|
| ## Training Details |
| - **Optimizer**: SGD with momentum 0.9, weight decay 1e-6. |
| - **Learning Rate**: Initial 0.01, reduced on plateau (factor 0.1, patience 10, min_lr 0.00001). |
| - **Data Augmentation**: Color jitter, random perspective, random horizontal flip, normalization. |
