This demo calculates the following types of visual similarity for apparel.
- Color Histogram
- Shape Similarity
- Pattern Similarity (two methods)
- Gabor features
- Local Binary patterns
Pattern similarity is calculated after classifying the pattern into either solid or pattern using a Linear SVM classifier. The demo has 926 images for women's apparel & 1193 images for footwear.
###Dependencies
- opencv (mac users https://jjyap.wordpress.com/2014/05/24/installing-opencv-2-4-9-on-mac-osx-with-python-support/ )
- scikit-image (http://scikit-image.org/download.html)
- django (https://www.djangoproject.com/download)
- If you want to build the index for every image ,you also need scikit-learn & numpy
###Running the demo
To start the demo navigate to visualsearch/src/search_webui
and start the webapp with this command
python manage.py runserver 8010 apparel
Runs the webapp locally on port 8010 for apparel
Similarly if you want to run the demo for footwear
python manage.py runserver 8011 footwear
Some things to know about the UI
- Refreshing the page shows a random set of 20 images
- Clicking on any image opens a popup that has three rows, each for a kind of similarity.
- The leftmost image is the query image (the one clicked)
- Only way to close the popup is to click on the X button in the top right
- The images in one row might overflow into another row depending on the width of the page :)
- By default three kinds of similarity are shown for every image: Color, Shape & Pattern(blend of gabor & LBP)
- To change defaults (number of random images, number of matches, kinds of similarities) edit
src/config.json
and restart webapp
#####Footwear screenshot
#####Apparel screenshot
##Questions
[email protected]