In recent years, there has been a notable increase in interest in image-based surface normal estimation. These approaches are capable of predicting the surface normal of real scenes using only an image, thereby facilitating a more profound comprehension of the actual scene and providing assistance with other perceptual tasks. However, dense regression predictions are susceptible to misdirection when encountering intricate details, which presents a paradoxical challenge for image-based surface normal estimation in reconciling detail and density. By introducing quaternion rotations as fusion module with geometric property, we propose a quaternion-based refined network structure that fuses detailed and structural information. Specifically, we design a high-resolution surface normal baseline with a streamlined structure, to extract fine-grained features while reducing the angular error in surface normal regression values caused by downsampling. Additionally, we propose a subtle angle loss function that prevents subtle changes from being overlooked without extra information, further enhancing the model's ability to learn detailed information. The proposed method demonstrates state-of-the-art performance compared to existing techniques on three real-world datasets comprising indoor and outdoor scenes. The results demonstrate the robust effectiveness of our deep learning approach that incorporates geometric prior guidance, highlighting improved robustness in applying deep learning methods. The source code will be released upon acceptance.

Coarse-to-fine schemes are widely used in traditional single-image motion deblur; however, in the context of deep learning, existing multi-scale algorithms not only require the use of complex modules for feature fusion of low-scale RGB images and deep semantics, but also manually generate low-resolution pairs of images that do not have sufficient confidence. In this work, we propose a multi-scale network based on single-input and multiple-outputs(SIMO) for motion deblurring. This simplifies the complexity of algorithms based on a coarse-to-fine scheme. To alleviate restoration defects impacting detail information brought about by using a multi-scale architecture, we combine the characteristics of real-world blurring trajectories with a learnable wavelet transform module to focus on the directional continuity and frequency features of the step-by-step transitions between blurred images to sharp images. In conclusion, we propose a multi-scale network with a learnable discrete wavelet transform (MLWNet), which exhibits state-of-the-art performance on multiple real-world deblurred datasets, in terms of both subjective and objective quality as well as computational efficiency. Our code is available on https://github.com/thqiu0419/MLWNet.