Detecting the Orientation of CelebA pictures using Deep Learning
This model has been trained on a modified version of the CelebA-faces dataset, which was made from flipping 20,000 images upside down and keeping 20,000 images intact.
The model relies on Resnet-18 as a backbone and is connected to one output node to classify whether the images are flipped upside down (1) or not (0).
Dataset used to train anisdismail/celebA-orientation-detection
Evaluation results
- Val F1 Score on CelebA-facesself-reported0.970