8000 which training stage the released code corresponds to · Issue #2 · Shua-Kang/TaKSIE · GitHub
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

which training stage the released code corresponds to #2

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitH 8000 ub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Open
JunQinwil1best opened this issue Jun 5, 2025 · 1 comment
Open

which training stage the released code corresponds to #2

JunQinwil1best opened this issue Jun 5, 2025 · 1 comment

Comments

@JunQinwil1best
Copy link
JunQinwil1best commented Jun 5, 2025 •
edited
Loading

Hi, thank you for sharing this great project — it's very helpful!

I noticed that in the original paper, the training of the generative model is divided into two stages:

Fine-tuning the original unconditional LDP model;

Training the conditional model using sub-goals.

I would like to confirm: does the training code provided in this repository only cover the second stage (training the conditional model)?
If so, was the first-stage fine-tuning of the unconditional LDP model done strictly following the official LDP training pipeline, or were there additional considerations?
@Shua-Kang
Copy link
Owner

Thank you for your question!

In my testing, training an unconditional diffusion model has helped ControlNet converge faster and achieve better results. For this part, you can simply use the original official code, as long as your dataset follows the required format.

You can refer to the original repository here:
🔗 https://github.com/huggingface/diffusers/tree/main/examples/text_to_image

When running the script, just set the dataset_name parameter to use this dataset:
🔗 https://huggingface.co/datasets/ShuaKang/calvin_d

@Shua-Kang Shua-Kang reopened this Jun 10, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@Shua-Kang @JunQinwil1best
0