The field of image generation and reconstruction is rapidly advancing with the development of diffusion models and flow matching techniques. Recent research has focused on improving the efficiency and quality of these models, particularly in the context of text-to-image synthesis, image-to-image translation, and medical imaging. One notable direction is the use of stochasticity and noise injection to enhance the performance of diffusion models and flow matching algorithms. Another area of research is the development of novel architectures and training methods, such as contrastive learning and score scaling sampling, to improve the fidelity and diversity of generated images. Noteworthy papers in this area include Smart-GRPO, which proposes a method for optimizing noise perturbations in flow-matching models, and Neon, which introduces a new learning method that leverages self-training to improve image generation quality. Overall, these advances have the potential to significantly impact applications such as image synthesis, data augmentation, and medical imaging analysis.
Advances in Diffusion Models and Flow Matching for Image Generation and Reconstruction
Sources
Smart-GRPO: Smartly Sampling Noise for Efficient RL of Flow-Matching Models
Fine-Tuning Diffusion Models via Intermediate Distribution Shaping
Universal Multi-Domain Translation via Diffusion Routers
Inference-Time Search using Side Information for Diffusion-based Image Reconstruction
Neon: Negative Extrapolation From Self-Training Improves Image Generation
Contrastive-SDE: Guiding Stochastic Differential Equations with Contrastive Learning for Unpaired Image-to-Image Translation
Diverse Text-to-Image Generation via Contrastive Noise Optimization
Joint Learning of Pose Regression and Denoising Diffusion with Score Scaling Sampling for Category-level 6D Pose Estimation
Diffusion Low Rank Hybrid Reconstruction for Sparse View Medical Imaging
TAG:Tangential Amplifying Guidance for Hallucination-Resistant Diffusion Sampling
Bidirectional Mammogram View Translation with Column-Aware and Implicit 3D Conditional Diffusion
Flow Matching for Conditional MRI-CT and CBCT-CT Image Synthesis
TDiff: Thermal Plug-And-Play Prior with Patch-Based Diffusion
Limited-Angle Tomography Reconstruction via Projector Guided 3D Diffusion
Three Forms of Stochastic Injection for Improved Distribution-to-Distribution Generative Modeling