what is this?

a scene classifier for 360 photos.

205 scenes are recognized including mountains, aquariums, bookstores, soccer fields, etc. exotic settings are included too - castles, chalets, and catacombs. top-1 accuracy is 50.04% and top-5 accuracy is 81.10%.

specs

• Ubuntu 14.04
• Caffe
• Numpy, SciPy, Pandas, Scikit Learn, Matplotlib

ml deets

trained on 2.5m images comprising 205 unique scene categories from mit's csail places. for details, read "places: an image database for deep scene understanding".

get da docker

building caffe from source is not for the faint of heart so you should do the following:

git clone https://github.com/mynameisvinn/scene_classification
cd scene_classification
docker build -t scene_classification .

run

option 1 (recommended)

from command line, do

docker run scene_classification python run_scene.py images/triple.jpg

you should see top five predictions, in order of confidence: water park, ocean, lagoon, beach, coast.

option 2

from command line, do

docker run -it mynameisvinn/scene_classification

then do

$ python run_scene.py images/park.jpg