Models based on diffusion has shown fantastic performance on image generation and other tasks. Although each day comes several papers about diffusion, there are several major topics about it :applications , theory and engineering improvement, conditional generation.
Year | Title | Venue | Paper | Code |
---|---|---|---|---|
2023 | Diffusion Models: A Comprehensive Survey of Methods and Applications | Arxiv | Link | / |
In this survey of diffusion models, more than 300 papers were surveyed, but only few months after it was published, another 300 papers has been produced,especially after Controlnet was released.
mainly about 3d diffusion applications and cross-modality applications, which are both very popular topics nowadays.
this part include text2image, few-shot , one-short and other researches about conditional generation. Generating images and videos under certain type of instructions has a prosperous future in commercial and daily uses. At the same time , this can be very difficult to satisfy people’s expectations.
Time | Title | Venue | Code |
---|---|---|---|
2022128 | Refining Generative Process with Discriminator Guidance in Score-based Diffusion Models | arxiv | link |
20210112 | D2C: Diffusion-Denoising Models for Few-shot Conditional Generation | arxiv | link |
20220623 | Entropy-driven Sampling and Training Scheme for Conditional Diffusion Generation | arxiv | link |
20220829 | Frido: Feature Pyramid Diffusion for Complex Scene Image Synthesis | arxiv | link |
20230221 | Diffusion Models and Semi-Supervised Learners Benefit Mutually with Few Labels | arxiv | link |
20211126 | Conditional Image Generation with Score-Based Diffusion Models | arxiv | link |
20230519 | Late-Constraint Diffusion Guidance for Controllable Image Synthesis | arxiv | link |
20230503 | Shap-E: Generating Conditional 3D Implicit Functions-openai | arxiv | [link](openai/shap-e: Generate 3D objects conditioned on text or images (github.com)) |
In fact, while diffusion models have a higher performance upbound, the training and inferring cost can be high. So it would be very interesting and meaningful to research about reducing unnecessary cost of diffusion-based models.
One of the most famous paper that fit this part is DDPM,which made application possible for diffusion models, but it is to old and famous too be presented in this chart.
Time | Title | Venue | Code |
---|---|---|---|
20220101 | Elucidating the Design Space of Diffusion-Based Generative Models | arxiv | link |
20210218 | Improved Denoising Diffusion Probabilistic Models-openai | arxiv | link |
2022 | Vector Quantized Diffusion Model for Text-to-Image Synthesis | CVPR | link |
some new dataset for diffusion models.
dataset | paper | github |
---|---|---|
Diversify Your Vision Datasets with Automatic Diffusion-Based Augmentation | link | link |
DiffuseExpand: Expanding dataset for 2D medical image segmentation using diffusion models | link | link |
Realistic Data Enrichment for Robust Image Segmentation in Histopathology | link | / |
A Multi-Institutional Open-Source Benchmark Dataset for Breast Cancer Clinical Decision Support using Synthetic Correlated Diffusion Imaging Data | link | |
Diffusion-based Data Augmentation for Skin Disease Classification: Impact Across Original Medical Datasets to Fully Synthetic Images | link | |
Diversify Your Vision Datasets with Automatic Diffusion-Based Augmentation | link | link |