home.social

#generative-model — Public Fediverse posts

Live and recent posts from across the Fediverse tagged #generative-model, aggregated by home.social.

fetched live
  1. Loopy, an AI model that generates lifelike portrait motion from audio input alone! 🎥✨ Loopy uses advanced temporal modules to sync natural movements like eye, eyebrow, and head motions with different audio types, whether it's speech, sighs, or even singing.
    #AI #DeepLearning #AudioDrivenAI #GenerativeModel #VideoSynthesis #Bytedance
    loopyavatar.github.io/

  2. Loopy, an AI model that generates lifelike portrait motion from audio input alone! 🎥✨ Loopy uses advanced temporal modules to sync natural movements like eye, eyebrow, and head motions with different audio types, whether it's speech, sighs, or even singing.
    #AI #DeepLearning #AudioDrivenAI #GenerativeModel #VideoSynthesis #Bytedance
    loopyavatar.github.io/

  3. Great post to follow along to learn diffusion model by implementing it from scratch. The distance function based perspective is new to me and makes a lot of sense with the visuals. chenyang.co/diffusion.html #Diffusion #GenerativeModel #FromScratch (via message.haoxiang.org)

  4. Distillation on top of k-rectified Flow makes a one-step SD called InstaFlow. Nice work. The reasoning makes sense to me, although I didn't fully get why it works. The overall workflow looks like a sophisticated distillation pipeline to me. Now this paper shows that it is possible to do it in one step, there must be some other way to get a similar linearly coupled target distribution. arxiv.org/abs/2309.06380 #Diffusion #GenerativeModel (via message.haoxiang.org)

  5. A nice way to model how one distribution transforms into another. The time-dependent vector field defines the flow and hence the probabilistic path across distributions. Interestingly diffusion model can be viewed as a special instance of this flow matching framework and one can alternatively define the path following optimal transportation theory as well. arxiv.org/abs/2210.02747 #Diffusion #GenerativeModel (via message.haoxiang.org)

  6. GZIP is all you need ...

    Language modeling is compression

    arxiv.org/abs/2309.10668

    "... Finally, we show that the prediction-compression equivalence allows us to use any compressor (like gzip) to build a conditional generative model."

    #compression #generativemodel

  7. GZIP is all you need ...

    Language modeling is compression

    arxiv.org/abs/2309.10668

    "... Finally, we show that the prediction-compression equivalence allows us to use any compressor (like gzip) to build a conditional generative model."

    #compression #generativemodel

  8. There will be two interesting #ComputationalNeuroscience seminars this Wednesday, both online:

    Simon Danner, "#ComputationalModels of spinal locomotor circuitry"
    ⏰ Wednesday, June 14, 2023 10:00 AM CET
    🌎 world-wide.org/seminar/9337/

    Neil Burgess, "Understanding our #memory system as a #GenerativeModel"
    ⏰ Wednesday, June 14, 2023 3:00 PM CET
    🌎 ucl.ac.uk/research/domains/eve

    #compneuro #neuroscience

  9. A #deeplearning model predicting unmeasured transcriptomic responses of single cells to single or combinatorial perturbations (drugs, gene knockouts etc.) ➡️ embopress.org/doi/full/10.1525
    from Fabian Theis @HelmholtzMunich
    #machinelearning #GenerativeModel #systemsbiology #singlecell

  10. A #deeplearning model predicting unmeasured transcriptomic responses of single cells to single or combinatorial perturbations (drugs, gene knockouts etc.) ➡️ embopress.org/doi/full/10.1525
    from Fabian Theis @HelmholtzMunich
    #machinelearning #GenerativeModel #systemsbiology #singlecell

  11. For anyone interested in AI-generated assets you might want to take a look at this paper from NVIDIA:

    "GET3D: A Generative Model of High Quality 3D Textured Shapes Learned from Images"

    nv-tlabs.github.io/GET3D/

    The code is available on GitHub:

    github.com/nv-tlabs/GET3D

    #AI #AIArt #3D #3DModel #GET3D #GenerativeModel #GameDev #NVIDIA

  12. For anyone interested in AI-generated assets you might want to take a look at this paper from NVIDIA:

    "GET3D: A Generative Model of High Quality 3D Textured Shapes Learned from Images"

    nv-tlabs.github.io/GET3D/

    The code is available on GitHub:

    github.com/nv-tlabs/GET3D

    #AI #AIArt #3D #3DModel #GET3D #GenerativeModel #GameDev #NVIDIA