metadata
language: en
tags: image-classification
license: mit
dataset: cifar10
metrics: accuracy (https://hf.co/metrics/accuracy)
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