Update README.md

README.md

@@ -1,3 +1,32 @@
 ---
-license: mit
+language : en
+tags     : cv
+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