Papers club from the AI team in D-ID - this time Diffusion Model(DM).
Diffusion Models were first introduced in Deep Unsupervised Learning using Nonequilibrium Thermodynamics. However, it took until Generative Modeling by Estimating Gradients of the Data Distribution (Song et al., 2019, Stanford University), and then Denoising Diffusion Probabilistic Models (Ho et al., 2020, Google Brain) who independently improved the approach.
A good explnantion on what are Diffusion Models and why they are intresting can be found in Diffusion-Models Tutorial (CVPR 2022).
מועדון קריאת מאמרים שלנו - כל ההרצאות בעיברית
| Title | Paper / Resource | Year | Why is it interesting? | Asignee | Recording | Slides |
|---|---|---|---|---|---|---|
| Denoising Diffusion Probabilistic Models | Denoising Diffusion Probabilistic Models | 2020 | read whyhigh quality image synthesis results using diffusion probabilistic models, a class of latent variable models inspired by considerations from nonequilibrium thermodynamics. |
@talbenha | zoom(@NnH10JK) | slides |
| The Annotated Diffusion Model | The Annotated Diffusion Model | read why |
self-work | -- | -- | |
| Colorization, Inpainting, Uncropping, and JPEG restoration | Palette: Image-to-Image Diffusion Models | 2021 | read whyA unified framework for image-to-image translation based on conditional diffusion models and evaluates this framework on four challenging image-to-image translation tasks, namely colorization, inpainting, uncropping, and JPEG restoration |
@ArnoBen | zoom (6CbWY6e*) | slides |
| Rethinking Diffusion Models Design | Elucidating the Design Space of Diffusion-Based Generative Models | 2020 | read whyKarras, the StyleGAN author is doing a back to the roots rethinking design choices of diffusion models, creating a well justified baseline archtecture |
@orgoro | zoom1(.m0gN7.?) zoom2(S^*c0ai3) | slides |
| Super-Resolution | Image Super-Resolution via Iterative Refinement | 2021 | read whyhigh quality image synthesis results using diffusion probabilistic models, a class of latent variable models inspired by considerations from nonequilibrium thermodynamics. |
self-work | -- | -- |
| Classifier (+ Classifier-Free) Diffusion Guidance | Diffusion Models Beat GANs on Image Synthesis & Classifier-Free Diffusion Guidance | 2021 | read whyDM achieve image sample quality superior to the current SOTA GAN models by improving the U-Net architecture, as well as introducing classifier (+calssifier free) guidance |
@talbenha | zoom(?JS330&C) | slides |
| Text2Image | ImageGen | 2022 | read whytext-to-image synthesis |
@alon.mengi | zoom(7hB61@CU) | slides |
| Efficient DM (Stable Diffusion) | High-Resolution Image Synthesis with Latent Diffusion Models | 2022 | read whyApply DM in the latent space of powerful pretrained autoencoders to enable training on limited computational resources while retaining their quality and flexibility |
@ShiraBaronn | zoom(U!+B+7g+) | slides |
| Imagic | Imagic: Text-Based Real Image Editing with Diffusion Models | 2022 | read whyApply complex (e.g., non-rigid) text-guided semantic edits to a single real image |
@Ganitk | zoom(%1x7WWl*) | slides |
| Text2Video | Imagen Video: High Definition Video Generation with Diffusion Models | 2022 | read whya text-conditional video generation system based on a cascade of video diffusion models |
@maysteinfeld | zoom($Y=U45cT) | slides |
| TTS-Diffusion | Grad-TTS: A Diffusion Probabilistic Model for Text-to-Speech | 2021 | read whyText-to-speech model with score-based decoder producing mel-spectrograms by gradually transforming noise predicted by encoder and aligned with text input by means of Monotonic Alignment Search. |
@amitay-nachmani | zoom(@3yMN0gC) | slides |
| 3D Shape Synthesis | LION: Latent Point Diffusion Models for 3D Shape Generation | 2022 | read whyHierarchical Latent Point Diffusion Model for 3D shape generation. LION is set up as a variational autoencoder (VAE) with a hierarchical latent space that combines a global shape latent representation with a point-structured latent space. |
@matan-feldman | zoom(q=v@4WYg) | slides |
| DreamFusion | DreamFusion: Text-to-3D using 2D Diffusion | 2022 | read whyDreamFusion use a pretrained 2D text-to-image diffusion model to perform text-to-3D synthesis |
@ShiraBaronn | zoom(9gZqV*2Y) | slides |
| FMRI-to-Image with SD | High-resolution image reconstruction with latent diffusion models from human brain activity | 2023 | read whyReconstruct images from FMRI using stable diffusion |
@Ganitk | zoom(B5J0vf?+) | slides |
| Few cool papers 😎 | Control Net, InstructPix2Pix, DreamBooth, Textual-Inversion, Prompt-to-Prompt | 2023 | read whyClosing the seminar with 5 cool papers |
@talbenha | zoom(r+52hd5@) | slides |