In this paper, we propose a learning-based depth estimation framework suitable for both densely and sparsely sampled light fields. The proposed consists of three processing steps: initial estimation, fusion with occlusion handling, refinement. can be performed from flexible subset input views. disparity estimates, relying on two warping error measures, allows us to have an accurate in occluded regions along the contours. contrast methods computation cost volumes, approach does not need any...

In this paper, we present a learning-based framework for light field view synthesis from subset of input views. Building upon light-weight optical flow estimation network to obtain depth maps, our method employs two reconstruction modules in pixel and feature domains respectively. For the pixel-wise reconstruction, occlusions are explicitly handled by disparity-dependent interpolation filter, whereas inpainting on disoccluded areas is learned convolutional layers. Due disparity...

This paper proposes a learning based solution to disparity (depth) estimation for either densely or sparsely sampled light fields. Disparity between stereo pairs among sparse subset of anchor views is first estimated by fine-tuned FlowNet 2.0 network adapted prediction task. These coarse estimates are fused exploiting the photo-consistency warping error, and refined Multi-view Stereo Refinement Network (MSRNet). The propagation from viewpoints towards other performed an occlusion-aware soft...

Light fields capture 3D scene information by recording light rays emitted from a at various orientations. They offer more immersive perception, compared with classic 2D images, but the cost of huge data volumes. In this paper, we design compact neural network representation for field compression task. same vein as deep image prior, takes randomly initialized noise input and is trained in supervised manner order to best reconstruct target Sub-Aperture Images (SAIs). The composed two types...

This paper presents JAWS, an optimization-driven approach that achieves the robust transfer of visual cinematic features from a reference in-the-wild video clip to newly generated clip. To this end, we rely on implicit-neural-representation (INR) in way compute shares same as We propose general formulation camera optimization problem INR computes extrinsic and intrinsic parameters well timing. By leveraging differentiability neural representations, can back-propagate our designed losses...

In this paper, we propose a novel light field compression method based on low rank-constrained neural scene representation. While most existing methods directly compress the views, our first learns Multi-Layer Perceptron (MLP)-based Neural Radiance Field (NeRF) from input views. To be able to efficiently NeRF representation, weights of MLP are optimized under low-rank constraint using Alternating Direction Method Multipliers (ADMM) optimization method. The then decomposed into Tensor Train...

Deep generative models have proven to be effective priors for solving a variety of image processing problems. However, the learning realistic priors, based on large number parameters, requires amount training data. It has been shown recently, with so-called deep prior (DIP), that randomly initialized neural networks can act as good without learning. In this paper, we propose model light fields, which is compact and does not require any data other than field itself. To show potential proposed...

We propose in this paper a Quantized Distilled Low-Rank Neural Radiance Field (QDLR-NeRF) representation for the task of light field compression. While existing compression methods encode set sub-aperture images, our proposed method learns an implicit scene form (NeRF), which also enables view synthesis. To reduce its size, model is first learned under (LR) constraint using Tensor Train (TT) decomposition within Alternating Direction Method Multipliers (ADMM) optimization framework. further...

In this paper, we propose a deep residual architecture that can be used both for synthesizing high quality angular views in light fields and temporal frames classical videos. The proposed framework consists of an optical flow estimator optimized view synthesis, trainable feature extractor convolutional network pixel feature-based reconstruction. Among these modules, the fine-tuning specifically synthesis task yields scene depth or motion information is well targeted problem. cooperation with...

How to explore useful features from images as prompts guide the deep image restoration models is an effective way solve restoration. In contrast mining spatial relations within prompt, which leads characteristics of different frequencies being neglected and further remaining subtle or undetectable artifacts in restored image, we develop a Frequency Prompting method, dubbed FPro, can effectively provide prompt components frequency perspective guild model address these differences....

Exploring motion information is important for the deblurring task. Recent window-based transformer approaches have achieved decent performance in image deblurring. Note that causing blurry results usually composed of translation and rotation movements window-shift operation Cartesian coordinate system by only directly explores orthogonal directions. Thus, these methods limitation modeling part. To alleviate this problem, we introduce polar coordinate-based transformer, which has angles...

In this paper, we leverage image complexity as a prior for refining segmentation features to achieve accurate real-time semantic segmentation. The design philosophy is based on the observation that different pixel regions within an exhibit varying levels of complexity, with higher complexities posing greater challenge We thus introduce guidance and propose Image Complexity prior-guided Feature Refinement Network (ICFRNet). This network aggregates both produce attention map Guided Attention...

The application of deep learning in traditional industries has not gained much attention. However, a great potential to be transplanted other fields. And we managed apply two techniques learning, object detection and tracking, dynamic counting. We test it on one the basic problem steel industry, rebar To cope with this, used an infrared camera collect video spot so that can distinguished from background apparently. Then use complete counting work. divided process into parts: tracking....

Unsupervised aspect identification is a challenging task in aspect-based sentiment analysis. Traditional topic models are usually used for this task, but they not appropriate short texts such as product reviews. In work, we propose an model based on vector reconstruction. A key of our that make connections between sentence vectors and multi-grained using fuzzy k-means membership function. Furthermore, to full use different representations space, reconstruct coarse-grained fine-grained...

The paper selects distributed solar water heating system of certain high-rise house in Jinan City as research object, its field performance test and energy efficiency assessment are made, obtains annual guarantee rate conventional substitution quantity the building, makeseconomic benefit assessment, analyzes saving effect economic system, gives out index index.Research results show building has good benefit. ForewordDesign, construction, management, etc. become increasingly normalized, but...

Light fields are a type of image data that capture both spatial and angular scene information by recording light rays emitted from different orientations. In this context, is defined as features remain static regardless perspectives, while refers to vary between viewpoints. We propose novel neural network that, design, can separate field. The represents using kernels shared among all Sub-Aperture Images (SAIs), sets for each SAI. To further improve the representation capability without...

Light field is a type of image data that captures the 3D scene information by recording light rays emitted from at various orientations. It offers more immersive perception than classic 2D images but cost huge volume. In this paper, we draw inspiration visual characteristics Sub-Aperture Images (SAIs) and design compact neural network representation for compression task. The backbone takes randomly initialized noise as input supervised on SAIs target field. composed two types complementary...

Axial light field resolution refers to the ability distinguish features at different depths by refocusing. The axial refocusing precision corresponds minimum distance in direction between two distinguishable planes. High can be essential for some applications like microscopy. In this paper, we propose a learning-based method extrapolate novel views from volumes of sheared epipolar plane images (EPIs). As extended numerical aperture (NA) classical imaging, extrapolated gives re-focused with...