Advances in Image Reconstruction and Generation

The field of image reconstruction and generation is moving towards more innovative and effective methods. Recent developments have focused on improving the fidelity and accuracy of image reconstruction, particularly in challenging scenarios such as dehazing and inversion attacks. Researchers are exploring new frameworks and techniques, including physics-grounded learning, momentum-based adaptive correction, and geometric-constrained approaches. These advancements have led to significant improvements in image quality and reconstruction accuracy, with potential applications in various fields. Notable papers include MAGIA, which introduces a novel label-inference-free framework for gradient inversion attacks, and HazeFlow, which proposes a physics-grounded learning approach for real-world dehazing. Additionally, Prompt-Guided Dual Latent Steering and Latent Iterative Refinement Flow have shown promising results in inversion problems and few-shot generation, respectively.

Sources

MAGIA: Sensing Per-Image Signals from Single-Round Averaged Gradients for Label-Inference-Free Gradient Inversion

HazeFlow: Revisit Haze Physical Model as ODE and Non-Homogeneous Haze Generation for Real-World Dehazing

Prompt-Guided Dual Latent Steering for Inversion Problems

Latent Iterative Refinement Flow: A Geometric-Constrained Approach for Few-Shot Generation

MeshMosaic: Scaling Artist Mesh Generation via Local-to-Global Assembly

Unleashing the Potential of the Semantic Latent Space in Diffusion Models for Image Dehazing

Generative Model Inversion Through the Lens of the Manifold Hypothesis

Built with on top of