Comments (6)
Thanks for the feedback. Based on our observation, our model peforms almost equally on small images and large images. We have to admit that given an input image and a casually-drawn mask, there are some possibilities that our method gonna fail, just as the failure examples shown in the supplementary materials. We appreciate the failure example you shared. We are working hard to improve the method further. Hopefully we will have some positive outputs in the near future.
from sample-imageinpainting-hifill.
@duxingren14 thanks for your reply. I further tested your algorithm on 256x256 resolution images, using some extremely thick masks, like this
input:
edgeconnect:
We found that in this case, edgeconnect performed better. After testing 1,000 test images, the EC psnr was 22.428, and your psnr was 18.702. Whether your algorithm needs to scale the tested image to 256 size, the mask area cannot be too dense. If so, then replace the inpainting module in your algorithm with edgeconnect, and then use the super-resolution method you designed. Will you get better results.
from sample-imageinpainting-hifill.
@ljjcoder, really appreciate the evaluation. We will do more evaluations on EdgeConnect and definitely try what you recommended.
from sample-imageinpainting-hifill.
Recently I did some more tests on both models. Here are some discoveries: 1. edgeconnect generally did better in inpainting edge structures, 2. EdgeConnect did better when mask percentage is over 30%,, while HiFill is prone to suffer "gray-strip" artifacts; 3, EdgeConnect did wrose in maintaining color consistency. It genuinely a good idea to merge the two to get a better model than either of the two.
from sample-imageinpainting-hifill.
The color consistency issue of EdgeConnect is shown as below, which is very typical for EdgeConnect.
input ____________________________ ______________ hifill _______________ _____________________ edgeconnect
from sample-imageinpainting-hifill.
The color consistency issue of EdgeConnect is shown as below, which is very typical for EdgeConnect.
input ____________________________ ______________ hifill _______________ _____________________ edgeconnect
I am very interested in your work on EdgeConnect+HiFill_SR. Would you like to open source or share this merged work with me?
from sample-imageinpainting-hifill.
Related Issues (20)
- It costs 6+s time to run post processing in your demo? HOT 12
- Attention Computing Branch architecture HOT 3
- Training code HOT 1
- How to train on 512*512 images with CRA? HOT 1
- Source of high resolution test samples HOT 1
- Convolution for mask and attention score training HOT 1
- InternalError (see above for traceback): cudnn PoolForward launch failed HOT 1
- Applying HIFILL to images smaller than 512 HOT 1
- attention score HOT 1
- How to do inference in batch? HOT 1
- When will the training code be open source? HOT 3
- Trouble on larger masks? HOT 4
- result is not good in other images
- result is not good in other images
- Is there anyway to run in 1.7x version Ascend ADK? HOT 1
- How to train my own data set? HOT 1
- tf2.x version
- most inpainting results suck
- Asking for help, I encountered a bug HOT 1
- pretrainned the model
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 sample-imageinpainting-hifill.