The image data of Natural Scenes around the world.Want to build powerful Neural network that can classify these images with more accuracy.
Link to the original problem statement on kaggle : https://www.kaggle.com/puneet6060/intel-image-classification
Files/directories used in the code:
categories :- jason file containig the 6 different classes ('buildings','forest','glacier','mountain','sea', 'street').
seg_train and seg_pred which are contained in a folder called classification which holds training and validation images.
seg_pred :- containing the images to be predicted.
I ran the Resnet 152 model and updated the pre-trained weights by training it on the train images dataset. Acheived an overall accuracy of 90.7% on the validation dataset (seg_test).
I have uploaded both an offline notebook as well as my google-colab notebook.