报告人：Prof.陈运锦，火箭军

报告题目：Trainable Nonlinear Reaction Diffusion (TNRD): A Flexible Framework for Fast and Effective Image Restoration

报告摘要：In this talk, I will discuss the widely known topic of nonlinear diffusion for image restoration. I will introduce one of our previous work, named TNRD -- Trainable Nonlinear Reaction Diffusion, which defines a flexible learning framework for various image restoration problems based on the concept of nonlinear reaction diffusion models. By embodying recent improvements in nonlinear diffusion models, we propose in our work a dynamic nonlinear reaction diffusion model with time-dependent parameters (i.e., linear filters and influence functions). In contrast to previous nonlinear diffusion models, all the parameters, including the filters and the influence functions, are simultaneously learned from training data through a loss based approach.

The TNRD approach is applicable for a variety of image restoration tasks by incorporating appropriate reaction force. We demonstrate its capabilities with a few applications, such as image denoising tasks with different noise types, single image super resolution, JPEG deblocking and image non-blind/blind deconvolution. Experiments show that our trained nonlinear diffusion models largely benefit from the training of the parameters and finally lead to surprisingly impressive performance with such a simple model. Our trained models preserve the structural simplicity of diffusion models and take only a small number of diffusion steps, thus are highly efficient. Moreover, they are also well-suited for parallel computation on GPUs, which makes the inference procedure extremely fast.