This paper proposes an end-to-end learning framework for multiview stereopsis. We term the network SurfaceNet. It takes a set of images and their corresponding camera parameters as input and directly infers the 3D model. The key advantage of the framework is that both photo-consistency as well geometric relations of the surface structure can be directly learned for the purpose of multiview stereopsis in an end-to-end fashion. SurfaceNet is a fully 3D convolutional network which is achieved by encoding the camera parameters together with the images in a 3D voxel representation. We evaluate SurfaceNet on the large-scale DTU benchmark.

We propose DeepHuman, an image-guided volume-tovolume translation CNN for 3D human reconstruction from a single RGB image. To reduce the ambiguities associated with the surface geometry reconstruction, even for the reconstruction of invisible areas, we propose and leverage a dense semantic representation generated from SMPL model as an additional input. One key feature of our network is that it fuses different scales of image features into the 3D space through volumetric feature transformation, which helps to recover accurate surface geometry. The visible surface details are further refined through a normal refinement network, which can be concatenated with the volume generation network using our proposed volumetric normal projection layer. We also contribute THuman, a 3D real-world human model dataset containing about 7000 models. The network is trained using training data generated from the dataset. Overall, due to the specific design of our network and the diversity in our dataset, our method enables 3D human model estimation given only a single image and outperforms state-of-the-art approaches.

Light field imaging has emerged as a technology allowing to capture richer visual information from our world. As opposed to traditional photography, which captures a 2D projection of the light in the scene integrating the angular domain, light fields collect radiance from rays in all directions, demultiplexing the angular information lost in conventional photography. On the one hand, this higher-dimensional representation of visual data offers powerful capabilities for scene understanding, and substantially improves the performance of traditional computer vision problems such as depth sensing, post-capture refocusing, segmentation, video stabilization, material classification, etc. On the other hand, the high-dimensionality of light fields also brings up new challenges in terms of data capture, data compression, content editing and display. Taking these two elements together, research in light field image processing has become increasingly popular in the computer vision, computer graphics and signal processing communities. In this article, we present a comprehensive overview and discussion of research in this field over the past 20 years. We focus on all aspects of light field image processing, including basic light field representation and theory, acquisition, super-resolution, depth estimation, compression, editing, processing algorithms for light field display, and computer vision applications of light field data. Index Terms-Light field imaging, light field processing.

The Reference-based Super-resolution (RefSR) super-resolves a low-resolution (LR) image given an external high-resolution (HR) reference image, where the reference image and LR image share similar viewpoint but with significant resolution gap (8×). Existing RefSR methods work in a cascaded way such as patch matching followed by synthesis pipeline with two independently defined objective functions, leading to the inter-patch misalignment, grid effect and inefficient optimization. To resolve these issues, we present CrossNet, an end-to-end and fullyconvolutional deep neural network using cross-scale warping. Our network contains image encoders, cross-scale warping layers, and fusion decoder: the encoder serves to extract multi-scale features from both the LR and the reference images; the cross-scale warping layers spatially aligns the reference feature map with the LR feature map; the decoder finally aggregates feature maps from both domains to synthesize the HR output. Using cross-scale warping, our network is able to perform spatial alignment at pixel-level in an end-to-end fashion, which improves the existing schemes [1,2] both in precision (around 2dB-4dB) and efficiency (more than 100 times faster).

No abstract