ObjectDetection with DETR from Facebook AI (FAIR)

This notebook is designed to be a simple introduction to Facebook AI's DETR model which can recognize different objects in an image. In this notebook, we do the following things:

  1. Use the Unsplash API to find images according to user-defined inputs
  2. Run the images through the DETR model in inference mode and return the top objects. This parameter can be defined by the user as well, but is set to 3 by default
  3. Using the outputs from the model, we trace bounding boxes and draw labels on the image

A sample selection of outputs is shown below

Demo

To-Do:

  • Return logits above a certain confidence threshold only
  • Find similarity between image tags and detected objects
  • Automatically add missing tags if confidence threshold is above a certain value