indoor-outdoor-image-classifier

A CNN-based i​mage classifier​ capable of detecting if a scene is ​indoors or outdoors.

This repository contains scripts for downloading videos corresponding to a particular or a few categories of youtube-8m dataset.

Dependencies

Python dependencies for the classification task

• Numpy
• Sklearn
• Tensorflow
• Keras
• tqdm
• PyYaml
• pytest

Dependencies for downloading youtube videos from ids

• youtube-dl

Dependencies for generation of frames from videos

• ffmpeg

Dependences for image preprocessing and resizing

• imagemagick

Install Python dependencies

 $ pip3 install -r requirements.txt

Test the pre-trained model

This CLI will run the pretrained model on a provided image.

bash classify.sh config/train_params.yml data/test/indoor_test.jpg

Download and extract the train/test data

An example dataset composed by 600 YouTube videos, for a total of 60.000 video frames, is availale here. The videos belong to different category, according to the youtube-8m dataset.

The dataset contains instances from the following categories, labeled as follows:

Living_room		indoor
Bedroom			indoor
Dining_room		indoor
Garden			outdoor
Outdoor_recreation	outdoor
Hiking			outdoor

Prepare the test environment downloading the provided dataset excerpt:

$ cd indoor-outdoor-image-classifier
$ wget http://insidecode.it/indoor-outdoor-data_64.zip
$ unzip indoor-outdoor-data_64.zip

Evaluate the model on the test split

This will test the performance of the model on the test split and run a simple unit test on two benchmark images.

$ bash evaluate.sh config/train_params.yml indoor-outdoor-data_64/frames

Train model

It is possible to tune the model parameters defining a new configuration file (following the default one in config/train_params.yml) and train the new model using the following command:

$ bash train.sh <config-file> <image-directory-path>