Pre Trained Diffusion Model
Pre-trained diffusion models are generative models used as powerful priors for solving various inverse problems, particularly in image processing and generation. Current research focuses on improving efficiency (e.g., one-step methods, faster sampling algorithms), enhancing control over generation (e.g., through guidance mechanisms and fine-tuning strategies like LoRA), and addressing security concerns (e.g., mitigating membership inference attacks). This work is significant because it leverages the strong generative capabilities of these models to achieve state-of-the-art results in diverse applications, ranging from image restoration and super-resolution to more complex tasks like image composition and 3D reconstruction.
185papers
Papers
March 26, 2025
March 23, 2025
March 20, 2025
March 11, 2025
March 10, 2025
March 3, 2025
February 18, 2025
February 10, 2025
February 5, 2025