Google Colab Notebook link: https://colab.research.google.com/drive/1iA8nvb93VLcrDfIt17AOIHnkVdLSNcW_?usp=sharing
This repo contains files for defining and creating a simple convolutional network for classifying/detecting the orientation of CIFAR-10 images (either normal orientation or flipped upside down/180 degrees). The following files are in this repo:
Coding_Challenge_for_Fatima_Fellowship.ipynb -- a copy of the Google Collab notebook with the code/output/writeup
best_model.pth -- dictionary of best model stats/weights found during training
cifar10flip_trn.pt -- saved training dataset of ~50% flipped CIFAR10 images
cifar10flip_tst.pt -- saved training dataset of ~50% flipped CIFAR10 images
image_examples.png -- an array of example imags from flipped CIFAR10 dataset
write-up -- write up of data processing, model results, and potential improvements (also in Google Colab)
wrong_predictions.zip -- a zip file of PNG images that were incorrectly classified by my model (each file name provide information on the image's prediction, true label, and its class)