File size: 964 Bytes
5075fbe 165f150 69aff1c 165f150 5075fbe 69aff1c 165f150 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 |
---
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 |