Model description

Upside down detector: Model to detect if images are upside down

• Picked a dataset of natural images - cifar10
• Synthetically turned some of images upside down. Created a training and test set.
• Trained it to classify image orientation ie if the image is upside down or not.

Intended uses & limitations

Intended to showcase skill set of being able to train a simple CNN classifier.

How to use

n/a

Limitations and bias

Trained on a relatively small dataset, hence it's hard to derive conclusions.

Training data

cifar10

Training procedure

Trained using Keras with Nadam classifier with ReduceLROnPlateau which halves the learning rate when the validation loss doesn't improve for 5 iterations

Evaluation results

The classifier was able to achieve 90% validation accuracy