littlewhitesea / stitchdiffusion Goto Github PK

This is the official implementation of "Customizing 360-Degree Panoramas through Text-to-Image Diffusion Models" (WACV2024)

Python 100.00%

stitchdiffusion's People

Contributors

Stargazers

Watchers

stitchdiffusion's Issues

Please tell how to run inference.

Colab notebook is not working. If you guys could please guide on how to generate images that would really help.
Thanks :)

About fine-tuned code and data sets

Hello, what a great idea. I learned from the paper that you used Panorama to fine-tune it. Could you share the fine-tuning code and data set? I would be very grateful.

colab implementation

Hi, thank you for your work.

Based on your provided code, I created a corresponding google colab (https://github.com/lshus/stitchdiffusion-colab).

Is this real equirectangular panorama image?

Hi, I was wondering whether the equirectangular projection is always ensured, or the finetuned model just know how to generate images seems like equirectangular patch? If equirectangular projection is ensured, it can be decomposed into multiple perspective view images right?

I have this question because I also tried similar things, to make Multi-Diffusion work with camera rotations. But I found the latents are very sensitive to warping operations. The results becomes terrible after several steps. I was wondering whether you have met the same problem? Do you find directly learning and generating equirectangular images helps to solve this problem?

Recommend Projects