Hello trend,
This is Satya Mallick from LearnOpenCV.com.
We have an exciting new post on Multi-Task Learning today on knee MRI dataset called MRNet
Before I explain the details, a quick reminder that EARLY BIRD pricing ($47) for OpenCV for Beginners ends in less 3 hours (10 AM Pacific Time, May 19)
With the help of 921 backers we have raised more than $106,274 so far.
MRNet – The Multi-Task Approach
In a previous post, we learned the details of the MRNet knee MRI dataset. It is a great post for beginners who have never worked on and do not know what an MRI dataset looks like.
In the same post, we also learned about MRNet challenge where our goal is to identify if the MRI shows one of the following diseases - abnormal, ACL tear, or meniscal tear. As a first stab at the problem, we trained 3 different models - one for each task.
Now, it is time to up our game. We will combine the three models and train a single model using an approach called Multi-Task Learning.
Multitask learning is inspired by human learning. When learning new tasks, we tend to apply the knowledge gained when learning related tasks. For example, the physical task of lifting weights also indirectly helps us build mental toughness and resolve that can be used for other tasks.
In today's post we will learn about a Mullti-Task approach called Hard Parameter Sharing. In this, you share the hidden layers between all tasks, while keeping several task-specific output layers. Without further ado, let's jump into the post and learn step by step how it is done
https://learnopencv.com/mrnet-multitask-approach/
and the code is at
https://github.com/spmallick/learnopencv/tree/master/MRnet-MultiTask-Approach
I hope today's post you learned something new and useful.
Also, please help us spread the word about our Kickstarter campaign. Tell your friends, family, and colleagues. Socially media posts help a lot too. These courses fund free material on our site.
OpenCV Weekly Webinar
In tomorrows OpenCV Weekly Webinar we will go over a tutorial for beginners.
Topic : Creating panoramas : Theory & Practice
Time : 9 AM Pacific Time, May 6, 2021
Hosts : Satya Mallick (CEO, OpenCV.org), and Phil Nelson (Content Manager, OpenCV.org)
Free registration : Zoom Registration Link
Also, check out last week's episode where Gerard Espona discusses his experience building 3D digital twins for OpenCV AI Competition 2021, and shows off some awesome Unity demos from their project: the open source Kauda robotic arm.
Cheers!
Satya
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.