Scene-Recognition

Demo of VGG-16 scene recognition model pre-trained on places-365 dataset.

Note : This is just a demo script to visualize quick results of scene contexts on images. The pre-trained models are all available in the MIT Places website.

Input-Output

Input

Image

Output

Top 5 scene contexts with the associated probability

Requirements

• python2
• Caffe
• Pandas

All codes are tested on a container built from Ubuntu 14.04 CPU/GPU docker image downloaded from floyd-hub(link given below).

Demo

1. Download the Scene_Recognition_models directory from Drive and place it in the same level as that of Scene.py
2. To find scene context of an image, run python Scene.py -i /full/path/to/image
3. To find scene contexts of all images inside a directory, run python Scene.py -d /path/to/directory. This will store the results in result.csv file

For testing the code, use the sample images provided in the example_images directory.

References

• The model/weight files are downloaded from Scene-365 Model zoo
• Docker Image
• MIT Places Dataset