MultiDiffusion icon indicating copy to clipboard operation
MultiDiffusion copied to clipboard

When will codes for region-based generation release?

Open Show-han opened this issue 2 years ago • 8 comments

Interesting work, and I wonder when will codes for region-based generation release. BTW, the hugging face demo for region-based generation crashed, can you fix it? I really want to try out

Show-han avatar Jul 01 '23 07:07 Show-han

the website is fixed now, I tried some examples on this website, but the results I obtained are significantly different from those presented in the paper. .......

企业微信截图_b4a81f02-ddfe-4968-b6cb-2d8aa5503352

hope to get some help from the authors.....

learningyan avatar Jul 03 '23 02:07 learningyan

the website is fixed now, I tried some examples on this website, but the results I obtained are significantly different from those presented in the paper. .......

企业微信截图_b4a81f02-ddfe-4968-b6cb-2d8aa5503352 hope to get some help from the authors.....

the website is fixed now, I tried some examples on this website, but the results I obtained are significantly different from those presented in the paper. .......

企业微信截图_b4a81f02-ddfe-4968-b6cb-2d8aa5503352 hope to get some help from the authors.....

I tried, and the results are significantly worse compared to those in the paper......

Show-han avatar Jul 03 '23 08:07 Show-han

Please note that by default, the bootstrapping parameter is set to 20, as in the application of tight masks. The high bootstrapping parameter encourages high-fidelity to accurate segmentation masks, but may not be suitable for rough bounding boxes (as in the provided example). The parameter can be controlled under 'advanced options'.

omerbt avatar Jul 05 '23 07:07 omerbt

hi Can you give me a set of detailed parameters to generate high-quality results? examples: prompts, bootstrapping parameter and so on . I have tried many experiments, but the results are bad ....

learningyan avatar Jul 05 '23 09:07 learningyan

Please note that by default, the bootstrapping parameter is set to 20, as in the application of tight masks. The high bootstrapping parameter encourages high-fidelity to accurate segmentation masks, but may not be suitable for rough bounding boxes (as in the provided example). The parameter can be controlled under 'advanced options'.

Could you give me an example of using this website to generate some good results?

learningyan avatar Jul 05 '23 09:07 learningyan

Please note that by default, the bootstrapping parameter is set to 20, as in the application of tight masks. The high bootstrapping parameter encourages high-fidelity to accurate segmentation masks, but may not be suitable for rough bounding boxes (as in the provided example). The parameter can be controlled under 'advanced options'.

I also tried almost every settings in adcanved options, but I found no one can generate images like the demos in your website.

Show-han avatar Jul 06 '23 04:07 Show-han

Requesting the authors to please upload sample inputs on hugging-face demo for region-control. I've tried everything from accurate segmentation masks to varying the bootstrapping param, but nothing seems to generate images anywhere near the likes of that are presented in the paper.

alphacoder01 avatar Aug 22 '23 13:08 alphacoder01

I have tried this example, with tiny modification, the result seems relatively good. image Any ideas when the author will release the codes?

uiyo avatar Aug 29 '23 03:08 uiyo