Trained models and Datasets for Indoor Scene Recognition via Object Detection and TF-IDF

Contained in this upload are the associated custom indoor object detection models (PyTorch); the object class lists for each model, label lookup files and the relative file locations for images used in the room prediction portion of the paper.

Object Detection model was trained using YOLOv5L. The github repository for which can be found at Ultralytics YOLOv5. 

We used a custom train/val/test split as places365 is an ongoing effort to improve scene classification and so they do not (or did not) provide labelled testing data.

We created a custom validation set from the training data and used the original validation dataset as the testing dataset.