AI Powered Search and Recommendation
Up to one-third of a company’s revenue can come from its
recommendations. Businesses recognize this fact and
invest a lot of money into building a good
recommendation engine. Nanonets APIs can help extract
products along with their attributes from an image which
is the fundamental step involved in recommendation
Detect and identify the type of furniture and its attributes from an image.
The user gets the ability to upload images of a room and get back similar furniture from the existing database. This not only serves as an intuitive way for the user to explore products but also allows recommending furniture while browsing.
We use cutting edge research in the deep learning field
to build accurate and scalable models. For this
furniture product recognition task, we built an
algorithm using a combination of the following 2 models:
1. Object Detection: This model identifies the location (bounding box around each furniture) and type of each furniture within an image. The model was trained: Sofa, Bed, Cupboard, Table, Side Table, Bookcase, Chair, and Desk
2. Multi-Label Classification: This model assigns tags/attributes to the furniture. The object detection is used to crop each detected furniture which is fed to this model to identify the furniture’s attributes like its material (wooden, metal), fabric type (leather, velvet, polyester), whether cushioned or not, color, how many seater (single seater, 2 seater) etc.
W150 images per type of furniture for the object detection model and 100 images per attribute for the multi-label classification model.
The model was deployed within customer’s premise on one
of their GPU enabled machine using a Docker image. The
machine is capable of processing each image within 40-50
milliseconds. Having the models running on your own
infrastructure means that your data never leaves your
premise which might be necessary depending on your
Note: Nanonets also provides hosted models. In this option, the models are hosted on Nanonets cloud and can be accessed through API calls over the web.
Building model (Fine tuning our pre-trained model on
your data): 2 days
Integrate model using Nanonets API: 2 days
- Enabled Search and Recommendation
- Increased user engagement
- Increased Revenue