Hello trend,
This is Satya Mallick from LearnOpenCV.com.
Today's post is the first prize winner of the recently concluded LearnOpenCV Blog Olympics.
You are probably familiar with the standard problem of object detection in computer vision. Given an input image the goal is to find an object like a person or a bike. The output of an object detector is an array of bounding boxes, and a class label associated with each bounding box.
Sometimes the object we are trying to detect is defined by its relationship to other objects in the image.
For example, a biker is a person on a bike. So, you can build a completely new object detector for "biker" if you detect a person, a bike, and have a way of formalizing the relationship "on."
Sometimes, two objects are identical visually and yet are different objects. For example, in a stick-shift car, the brake pedal and the clutch pedal are visually identical. The only way we can tell them apart is by looking at them together - the brake pedal is to the right of the clutch pedal.
Such relationships can be represented using graphs.
At other times, we can build special detectors by combining multiple attributes. For example, we can train a cup detector, and separately train a red color detector and combine the two attributes to build a detector for red cups.
We will explore the idea of using multiple attributes and combing relationships between objects using graphs in today's post.
https://learnopencv.com/multi-attribute-and-graph-based-object-detection
The code is shared within the post.
Computer Vision & AI Predictions for 2022.
A couple of weeks back, I discussed a few predictions for Computer Vision and AI on the OpenCV Weekly Webinar with Joseph Nelson, the CEO of Roborlow. Click on the link below for a recording of the webinar.
If you like the content, please subscribe (zoom link) to OpenCV Weekly Webinar.
Cheers!
Satya
Courses / Webinar / Jobs
Facebook / LinkedIn / Twitter / Instagram
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.