Comments (7)
I think @bongbonglemon refers to B', which is defined in the paper as the perturbed version of the true background B. Correct me if I'm wrong.
I'm interested in your work, so sometimes I take a look at your GitHub page. :)
from background-matting.
We can write the captured image I=\alpha*F + (1-\alpha)*B, where B is the true background. Now we capture background when the subject leaves the scene. This background is not the true background. Why is it not the true background? Because there are small changes w.r.t. shadows, reflections, camera resampling etc. Thus we call this captured background as B'.
We train on synthetic-composite Adobe dataset, which provides us with image, alpha matte, and foreground (I, F,\alpha). We can easily calculate the true background B. But we do not know how the captured background B would look like. Thus we perturb B to obtain B' by adding sample gaussian noise, gamma correction etc.
from background-matting.
Pre-processing script for testing only does perspective alignment and bias-gain correction. Perturbing B' is only needed for training.
Given the foreground segmentation mask, m, dilate the mask up to certain steps, say cv2.dilate() with 100 steps to create a new mask m1. Add gaussian noise in the (m1-m) region. It may not be super important, as long as you add some Gaussian noise with random mu, sigma and some gamma change it is fine.
from background-matting.
Can you be a bit more specific about the "perturb version" of B? Sorry, I can not remember where I used this term, and for what context.
The background images in sample data are raw backgrounds. You have to align it with our data based on the provided code.
from background-matting.
@senguptaumd Yes I understand why you did it. I was asking as to where the code was (in this repository) for the "adding sample gaussian noise, gamma correction etc."
from background-matting.
Sorry, I do not have the training code ready yet. It will take some time. But you can follow what is mentioned in the paper. I do plan to release the training code on synthetic and real data shortly.
from background-matting.
Alright. Yeah I saw that you mentioned that in the README but I was half guessing that it should be in the preprocessing script. And yes I would like to implement it myself. I'm just a bit confused by what you mean by adding the gaussian noice and gamma correction "around the foreground region". Does that refer to the area that the subject was covering? And does "around" refer to the just the perimeter or is it the whole area? Thank you for time!
from background-matting.
Related Issues (20)
- MSCOCO HOT 2
- Can I skip the Alignment step? and How? HOT 1
- issue in JPG inputs HOT 1
- loss log HOT 1
- About the training time HOT 1
- matting performance problem HOT 1
- Real time results with constant background HOT 1
- SAD in the paper
- Evaluation on Synthetic-Composite Adobe Dataset
- [error] Could not create cudnn handle: CUDNN_STATUS_INTERNAL_ERROR HOT 1
- Background Image HOT 3
- How to test after training on unlabeled real videos
- Camera Model For Fix-Camera dataset
- 这个对抗损失的意义是什么?怎么理解? HOT 1
- The download link for pre-trained model is failed! HOT 1
- Possible Memory Leak with AdobeDataAffineHR
- 是否可以训练他来分割别的东西,比如说超市商品?
- Could he be trained to split other things, such as supermarket goods?
- Using sample data does not achieve the effect in the paper
- Colab notebook is no longer works HOT 1
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 background-matting.