The dataset we’ll be using for this case study is the Indoor Scene Recognition (also called Indoor CVPR) dataset released by MIT. This database contains 67 indoor categories room/scene categories, including homes, offices, public spaces, stores, and many more.

You will find dataset here

After downloading dataset unzip it and put images folder to indoor_cvpr directory

1. create_dataset.py: This file is used to build the training and testing sets for our input dataset
2. extract_features.py: This file is used to HDF5 file for the dataset splits
3. train_model.py: This file is used to train a Logistic Regression classifier to recognize image orietations and save the resulting model in the models directory
4. orient_images.py: This file is used to apply orient to testing input images

To run create_dataset.py, execute following commmand

python create_dataset.py --dataset indoor_cvpr/images --output indoor_cvpr/rotated_images

After that you will encounter 4 folders inside indoor_cvpr/rotated_images directory named 0, 90, 180 and 270.

To run extract_features.py, execute following commmand

python extract_features.py --dataset indoor_cvpr/rotated_images --output indoor_cvpr/hdf5/orientation_features.hdf5

After that you will encounter orientation_features.hdf5 inside indoor_cvpr/hdf5 directory

To run train_model.py, execute following commmand

python train_model.py --db indoor_cvpr/hdf5/orientation_features.hdf5 --model models/orientation.cpickle

To run orient_images.py, execute following commmand

python orient_images.py --db indoor_cvpr/hdf5/orientation_features.hdf5 --dataset indoor_cvpr/rotated_images --model models/orientation.cpickle

After that you will encounter orientation.cpickle inside models directory

[INFO] best hyperparameters: {'C': 0.01}
[INFO] evaluating...
              precision    recall  f1-score   support

           0       0.94      0.93      0.94       627
         180       0.91      0.93      0.92       633
         270       0.90      0.90      0.90       621
          90       0.90      0.90      0.90       614

    accuracy                           0.91      2495
   macro avg       0.91      0.91      0.91      2495
weighted avg       0.91      0.91      0.91      2495

[INFO] saving model...