Cold-Diffusion-Models VS VideoX

Sure you can. You might be interested in cold diffusion (https://arxiv.org/abs/2208.09392) which tries doing a bunch of different kinds of degradation processes besides adding gaussian noise. You can kind of choose whatever input corruption process you want and teach the model to reverse it, and it works kinda well (I think gaussian noise might be better though)

I wrote a response yesterday but did not post it or send it, ops.

I still don't understand the problem, if you ask model trained on a noise pattern "trees" for a forest it will still give you a random forest, that's what it was trained on, also: https://arxiv.org/abs/2208.09392, to see the diffusion process applied to processes other than Gaussian noise.

What about this research: https://arxiv.org/abs/2208.09392

Found relevant code at https://github.com/arpitbansal297/Cold-Diffusion-Models + all code implementations here

This claims to explain diffusion models from first principles, but the issue with explaining how they work is we don't know how they work.

The explanation in the original paper turns out not to be true; you can get rid of most of their assumptions and it still works: https://arxiv.org/abs/2208.09392

Check out the cold-diffusion repo, which has nice clean implementations, and also is useful in pointing out that the multi-step computation idea isn’t limited to denoising. https://github.com/arpitbansal297/Cold-Diffusion-Models

Input Output Theme Use Case Example Platforms Text Text Marketing Copywriting, creative personalisation, SEO optimisation Jasper AI (https://www.jasper.ai/) WRITER (https://writer.com/) Video Video / Text Sales Business development agumentation, Sales coaching https://www.regie.ai/ https://www.oliv.a/ Audio Text / Audio CRM Customer service chatbots (answering tickets and queries) https://symbl.ai/ https://forethought.ai/ Audio / Text Audio / Text Talent Management Performance management, Job interviews, Coaching, training https://www.onloop.com/ https://www.converzai.com/ Text Text Legal Contract drafting, Legal validation / citations https://casetext.com/ Text Text Code creation Natural language to generate code for softwares. AI pair programmer https://mayalabs.io/ https://github.com/features/copilot https://debuild.app/ https://kombai.io/ Text Text Code Documentation Natural language to generate code for softwares. AI pair programmer https://mayalabs.io/ https://github.com/features/copilot Text Image Art Generation AI system that can create realistic images and art from a description in natural language. https://openai.com/dall-e-2/ https://midjourney.com/home/?callbackUrl=%2Fapp%2F Text Text ML Model Build, train and deploy AI models https://huggingface.co/ https://replicate.com/ Text Audio Voice Synthesis Educational platform, training, Cognitive coaching https://synthesys.io/ https://speechify.com/ Text Video Video Educational platform, training, Cognitive coaching https://www.synthesia.io/ https://github.com/microsoft/VideoX/tree/master/X-CLIP Text Image 3D modelling Storyboarding for games, 3D video modelling https://nv-tlabs.github.io/GET3D/ https://dreamfusion3d.github.io/

►Code: https://github.com/microsoft/VideoX/tree/master/X-CLIP

The pieces are coming into place https://github.com/microsoft/VideoX/tree/master/X-CLIP

Continue reading | Check out the paper and github link