Comments (5)
Hi, thanks for this great question, and I think there are two potential considerations for this:
- Keep the consistency between training and inference over all tasks. The model samples from random noise and y_cond in the inference stage.
- y_cond can distinguish between the mask and unmasked areas since y_t may not be straightforward enough when t is small.
from palette-image-to-image-diffusion-models.
Hi, thanks for this great question, and I think there are two potential considerations for this:
- Keep the consistency between training and inference over all tasks. The model samples from random noise and y_cond in the inference stage.
- y_cond can distinguish between the mask and unmasked areas since y_t may not be straightforward enough when t is small.
Thanks for the kind reply. This makes sense, though it is worth trying the second reason. I will post here if I realized something different.
from palette-image-to-image-diffusion-models.
Feel free to reopen the issue if there is any question.
from palette-image-to-image-diffusion-models.
@PouriaRouzrokh I have opened a separate issue on it. But I am in urgent need of a solution, so I just wanted to check with you .
In my inpainting case, during the inference only the y_cond
and mask
images are given. In that case, may I know how to do a inference?
In the network.py
script, for the inpainting task the below line will be executed as part of the restoration
function. As y_0
is None for me, I am not sure how to deal with this line. If I skip the below line then the results are very bad (just only some whitish kind of image is generated). Also, in the Process.png
image I can notice that for each step the noise level is increasing rather than decreasing.
if mask is not None:
y_t = y_0*(1.-mask) + mask*y_t
Any idea on how to proceed?
from palette-image-to-image-diffusion-models.
@Janspiry Why not just set the mask as y_cond
? for consistency among all tasks?
from palette-image-to-image-diffusion-models.
Related Issues (20)
- hello
- Inputs and Outputs of Different Sizes HOT 1
- Question about encode the gama rather than t HOT 2
- pth2onnx,How should I use βtorch.onnx.export()β HOT 1
- Image-to-image translation with mostly black images HOT 3
- How can I add classifier guidance while doing the uncropping task?
- Broken pipeline error while training on multiple gpu
- use this project for image restoration
- How can I adapt the colorization model to work with different image resolutions?
- Training loss growing up
- why p_mean_variance use noise_level instead of sample_gammas like in training for time conditon of denoise function. HOT 3
- There was no result at the time of the test
- segmentation fault HOT 1
- Some of the results are full of noise. HOT 2
- test noise schedule and train noise schedule are different?
- Whether to use a lr scheduler when training from the scratch? HOT 1
- I'm fused by the output and target noise.
- [Uncropping]How to generate panoramas like Firgure 2?
- Error During Colorization Training
- How to implement JPEG restoration task based on this paper?
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
π 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. πππ
-
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 β€οΈ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from palette-image-to-image-diffusion-models.