Comments (2)
Hi @nttung1110, thanks for your interest!
The intuition behind this choice is that, as mentioned in the paper, most of the spatial information (e.g., location of objects) is determined in the first denoising steps, while the last ones mainly influence fine details. Since our goal is to encourage the generation of neglected subjects, we perform larger step sizes in the first denoising steps.
You can also see the impact of applying our method in the last timesteps in Fig. 12 of the paper. The figure demonstrates that these steps mostly add artifacts and do not induce meaningful changes in the resulting image.
Referring to your specific issue- I do not recall a convergence issue with our method. The main issue that I saw before adapting the step size was that the loss decreased very significantly but resulted in OOD latents and OOD images.
Without further details on the objective you are exploring it's hard to come up with an explanation to why this is happening in your case.
I would start with very large step sizes just to make sure that your loss can decrease (without our method).
I hope this helps!
from attend-and-excite.
Hi,
Thanks for your detailed explanation. I would try to explore more to fix this issue.
from attend-and-excite.
Related Issues (20)
- Dreambooth and working with new tokens HOT 8
- Can this be used with the automatic1111 webui? HOT 6
- KeyError:'up' when generating images HOT 2
- Licensing? HOT 1
- How about global properties? HOT 2
- How do I give the token indices if I want to emphasize a group of words? HOT 2
- WebUI? HOT 1
- Explainabililty HOT 1
- Prompts for the quantitative results HOT 2
- "a spiky porcupine", what should I do for procupine, which contains three tokens HOT 1
- How to visualize the cross-attention map of each token HOT 2
- About the Memory usage HOT 2
- some issuses about stable-diffusion-2-1-base HOT 1
- perhaps you can be an add-in for sdwebui? HOT 1
- Did you release the evaluation dataset in your paper? HOT 2
- Comfyui HOT 3
- Latent not updated on first iteration of iterative refinement HOT 4
- KeyError: 'up_cross' HOT 2
- Are the datasets(Animal-Animal Animal-Object Object-Object) avaliable ? 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 attend-and-excite.