/Indoor-Scene-Understanding

Computer Vision project for the analysis of indoor house scenes including a furniture retrieval system.

Primary LanguagePython

Indoor-Scene-Understanding

Authors: Giuseppe Cartella, Sara Sarto, Kevin Marchesini

Computer Vision project for the analysis of indoor house scenes including a furniture retrieval system.

Preliminary steps

  1. Download from drive (request access to authors) the folder containing all trained models and dataset.

  2. Copy all dataset folders to /retrieval folder of the project.

  3. Move 'model_mask_default.pt' and 'model_mask_modified.pt' to /Indoor-Scene-Understanding

  4. Move 'dataset_embedding.pt' to /Indoor-Scene-Understanding/retrieval/method_autoencoder

  5. Move 'descriptors.pkl' to /Indoor-Scene-Understanding/retrieval/method_SIFT

  6. Move 'MLP_model.pt', 'randomforest_model.pt' and 'dataset_all_objects.csv' to /Indoor-Scene-Understanding/classification

  7. Install the requirements.txt in your virtual environment

pip install -r requirements.txt

Execute the pipeline

In the root folder execute:

python execute_pipeline.py -img test_images/bedroom.jpg -mdl modified -rtv autoencoder -clf forest

Available options mdl : ['default', 'modified'], rtv : ['sift','dhash','autoencoder'], -clf : ['forest', 'mlp']