Comments (5)
Haussdorf loss is a distance based loss so it doesn't have upper boundary of values it can take - it is solely based on the dataset itself. I don't think anything is wrong here just a way the loss behaves - raising distance fields to power of alpha generates such big values but it's in order with the paper itself. May I ask, did you expect it to be between 0-1 thus your confusion?
from hausdorffloss.
Thank you for your reply.
I want to combine HD Loss and Dice Loss together.
However, the Dice Loss is between 0-1.
So I want to normalize it into 0-1.
from hausdorffloss.
Problem with normalizing Haussdorf itself is that its max value can change between batches thus simple normalization technic would be batch dependent which it's not correct. Maybe you could try to find max haussdorf value over whole dataset and use it as a normalization term to divide by. Though I don't know how this would behave.
Common idea for using hybrid losses is to introduce some parameter beta:
loss = (1-beta) * dice + beta * HD
Selection and usage of such parameter is described in this paper: https://arxiv.org/pdf/1812.07032.pdf. It's about boundary loss but it is also a distance based loss so it should work fine as described in this paper in experiments section https://openreview.net/pdf?id=NDEmtyb4cXu. From my experience starting with bigger parameter for region based losses and decreasing it over time to amplify distance loss term is a way to go, cause distance losses are a bit unstable when predictions are too random, but it may depend on data.
Hope you find those ideas useful in your problem.
from hausdorffloss.
Wow~ your idea is amazing! Thanks for your help.
from hausdorffloss.
Thank you for your reply. I want to combine HD Loss and Dice Loss together. However, the Dice Loss is between 0-1. So I want to normalize it into 0-1.
Hi, @ZhixiangWang-CN
Have you normalized Hausdorff Loss to 0-1?I want to combine HD Loss and Dice Loss together too,but I don't know how I can solve this problem. Could you give me some help ?
Thanks in advance.
Looking forward to your reply.
Best.
from hausdorffloss.
Related Issues (6)
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from hausdorffloss.