Dongliang Chang

Associate Researcher@BUPT | Opinions are personal

[Dongliang Chang @DL_Chang]

@dasayan05 hahahahaha

[CVSSP Research @cvssp_research]

Huge congratulations to PAI and CVSSP colleagues whose papers have been accepted for the Conference on Computer Vision and Pattern Recognition #CVPR2024 🤩
See their papers here 👉https://t.co/YFCc6a8D7m

[Dongliang Chang @DL_Chang]

🌕

[Kam Woh Ng @kam_woh]

🐦🦆🦅🦉Excited to share #DreamCreature, which creates novel creatures by learning *1000+* fine-grained sub-concepts.

🚨Code & Gradio demo on Colab is available now!!

⌨️GitHub: https://t.co/dfrrcEFrRa
📜paper: https://t.co/PR6MStWU7r

[Dongliang Chang @DL_Chang]

🤫

[University of Surrey @UniOfSurrey]

We are feeling very festive today! For campus opening hours over the winter break, visit:
https://t.co/6RxWUFrxdC

[Dongliang Chang @DL_Chang]

⛄️⛄️❄️❄️

[Kam Woh Ng @kam_woh]

@krishnanrohit @ysu_nlp To create new animals that do not exist, we proposed a solution called DreamCreature, which learns to "draw" the parts of the animal (in our paper, we show bird & dog only) with SD and recombine them to create new animals.

https://t.co/yOUFVjtwom

[Hindustan Times Tech @HTTech]

University of Surrey introduces DemoFusion, a game-changing AI tool, democratising high-res image creation without costly subscriptions or powerful computers.

https://t.co/mEZEfTYSxb

[Yulia Gryaditskaya @ygryaditskaya]

Fully funded PhD studentships for outstanding Chinese students at CVSSP&PAI, UK: https://t.co/y22g3QD8Co.
Start date: 1 October 2024
Duration: 4 years
Application deadline: 12 January 2024
About CVSSP: https://t.co/EypGfR5UE1
About PAI: https://t.co/eBKOYGvFv0

[Dongliang Chang @DL_Chang]

🎉 Exciting News from Our Team! 🌟

We're thrilled to announce a major milestone for our open-source project, DemoFusion: We've just surpassed 1,000 stars on GitHub! 🚀

Check out the project here: https://t.co/leGa3LbFfo

#DemoFusion #OpenSource #SketchX

[𝐑𝐢𝐤𝐤𝐚𝐫.tez.eth | #omencollective | #MAIF @socalpathy]

Pretty interesting "upscaling" results with DemoFusion.

Started with 1024x1024, and ended up at 3072x3072. I'll drop samples in the thread to show how each image turned out + the repo at the end to check out.

Here is the starting image 🧵

[Subject News @subject_news]

DemoFusion is a new tool made by the University of Surrey that promises to bring low-cost, high-resolution artificial intelligence (AI) image generation to a wider audience without the need for powerf…

https://t.co/oU708P0sgg
#ai #image #demofusion #high-resolution #powerful

[AI 托我给您带个话 @AIWhispers4U]

🔗 https://t.co/ArXmGqJZ0g

DemoFusion：无需 $$$ 即可实现高分辨率图像生成的民主化 🖼 Img2Img Colab 🥳感谢 @RuoyiDu ❤ @DL_Chang ❤ @tmh31 ❤ @yizhe_song ❤ Zhanyu Ma ❤

🌐页面：https://t.co/tFvuOWsvBQ

📄论文：https://t.co/FBYwQLUN4b

🧬代码：https://t.co/UBVNvf3ggR

🦒colab 由 https://t.co/68zTQtje7O 提供：请尝试 🐣 https://t.co/J07hRAa5jH

[Lyson Ober @lyson_ober]

正巧昨晚 @0xstory_xhu 也和我聊到这个，实测后我发现有些摄影场景一点也不比 @Magnific_AI 差，真是尴尬。我主观上感觉 Magnific 在动漫场景的细节上更细腻，明天我发发测试图。大家可以看看小互的详细介绍先。相比较于 Magnific 昂贵的价格，DemoFusion 还是免费的…

[Wanheda @Wanheda]

DemoFusion: Democratizing the Generation of High-Resolution Images 🌟

We have seen image enhancement tools (upscaling) with high prices, but now DemoFusion arrives, changing the game.

This innovative tool is democratizing the generation of high-resolution images, making it accessible to everyone, and best of all, it's free!

Today, we will explore how DemoFusion is opening up a new world of possibilities in digital art and creativity.

Thanks for research @ruoyidu's and also thanks to @radamar for sharing 🫶🏻

Have you already experimented with DemoFusion?

https://t.co/59cJFEJpIq…

[Gaurav Aggarwal @GAggarwal]

@annabella23521 Thanks! I use opensource diffusion models to generate pics generally for fun purposes - the above one is generated using opensource demofusion - https://t.co/zQnBlDty59

[weir-欧sir @weiou919311215]

清华大学新发布的DemoFusion这个模型简直是太棒了！！

--人性化：主打民主化高分辨率图像生成，完全免费！

--超高分辨率放大：可将SDXL生成的1024×1024分辨率图像，提升至4倍、16倍，甚至更高！

--低内存资源，开箱即用：DemoFusion设计得既高效又易于使用，不需要大量的内存资源或复杂的调整过程！

[hamyori @hamyori]

demofusion xl

[歸藏(guizang.ai) @op7418]

尝试了一下DemoFusion这个图片放大项目，这玩意的资源消耗真是恐怖，2048*2048的图片A100居然跑了两分半的时间。
前两张是动漫图片，后两张是写实照片，不过效果确实好。可能是实现问题部分细节有一些崩坏的纹理。
动漫的图片会被识别成写实的照片，可能是提示词问题。
原图放在下面的推里面了可以对比一下生成前后的效果。

想体验的可以用这个Colab链接尝试，不过不是Pro的就算了，T4得跑好久：https://t.co/UC7BNA69V8