Skip to content
Sean Bettsv2.0.2
Back to Building
AI prototypeArchived

AI Brand Detection

An AI prototype for detecting brands and products in video, exploring how to automate the measurement of product placement and sports sponsorship.

Sole developer · August–December 2024View on GitHub (opens in a new tab)
Original Brand Detection output showing a red detection box labelled pizza hut around the logo on a pizza box in a film frame.
In practice

What it does

  1. Detect brands and products in video frames.
  2. Identify multiple brands within the same video.
  3. Mark detections with labelled bounding boxes so the results can be inspected.
Behind the work

The challenge

Product placement and sports sponsorship deals still rely heavily on manual measurement. I built this prototype to explore whether the latest AI models could automate that work by detecting brands and products in video. The challenge was recognising them across different scenes, lighting conditions and contexts, while making the process economical enough to be useful commercially.

What I learned

I could build the detection system, but I could not make it commercially viable at the scale I was operating at. The project made the distinction between technical capability and a sustainable product very concrete. Producing useful detections was one part of the problem; delivering them at a viable cost was the constraint I could not resolve.

Built with

  • React
  • Python
  • PyTorch
  • TheFuzz
  • FastAPI
  • GCP
  • Google Cloud Speech API
  • Google Cloud Vision API
  • Segment Anything Model 2