While watching some videos we often come across some products which catches our eyes buy we usually miss on them. That's a lot of missed opportunities for the company to generate revenue of their sponsored products. And we also don't want to out of the blue, irrelevant ads.
What it does
While you are watching your content on the OTT service, whenever a product comes on the screen a toast subtly slides on the screen with 2 options. Either you can go and check that product on the website right then, the video will pause and you'll be redirected to the link where you can buy it. The second options enable you to quietly add that product to your wishlist which you can review later.
How we built it
We trained our model with a set of products. The video was taken at 1 FPS (hardware limitation) and it was run through the model. The objects were identified and a JSON object with product details (link and name) and the timeframe is sent to the device where the ad is displayed accordingly.
Challenges we ran into
It was particularly hard to design an ad in such a customized and non-distracting way. Also, to train the model to identify the objects in a video with limited time and hardware was a challenge.
Accomplishments that we're proud of
The ads were very subtle and interactive and not disturbing the user in any way. We were able to preserve the user experience while increasing the customizability of the ads. We trained our model on 1FPS and we were able to achieve quite accurate and real-time results within the given timeframe.
The ads to click ratio on OTT platforms is really low. Plus the ads on these platforms are really annoying and hampers user's experience. Ads are somewhat related but still not much customized. With this model, we will be able to generate a high ratio of ad to click ratio and generate organic leads for the products. It'll encourage more brand-based companies to endorse the content creators on the platform which will generate 2-way profit for our platform.
What's next for Watch and Buy
Next is high-level user profiling and cross-platform integration. More variety of products and diverse database for powerful object recognition. Real-time analysis of dynamic content.