Zero-shot learning (ZSL) utilizes semantic information that is auxiliary information to transfer knowledge from seen classes to unseen classes, thereby realizing the recognition of unseen classes. The generative metho...
详细信息
In many application fields, an appropriate high-quality fast image upsampling method is required. Although many interpolation-based upsampling methods have been proposed, the quality of result images is not satisfacto...
详细信息
In many application fields, an appropriate high-quality fast image upsampling method is required. Although many interpolation-based upsampling methods have been proposed, the quality of result images is not satisfactory. Some of them are very fast, but produce poor quality images, the others can produce high quality images, but the methods in them are slow. In our paper, we proposed a fast statistical image upsampling method based on CUDA, it can obtain high quality images based on reducing the input resolution-grids dependency artifacts. Thus, we can rebuild low resolution images' sharp edges fast and get high-quality upsampled images in real time. We have applied this method in the multi-resolution texture generation of large scale terrain rendering. Experiments prove that our method can receive ideal effects in real time.
In this paper, we propose a novel approach for image completion with automatic structure propagation. This method integrates two stages: Firstly, it extends the salient structure lines from the known regions to the un...
详细信息
In this paper, we propose a novel approach for image completion with automatic structure propagation. This method integrates two stages: Firstly, it extends the salient structure lines from the known regions to the unknown by following a local self-similarity assumption on natural images. Then guided by the structure information, it restores the missing region by patch-based texture synthesis. Experiment results demonstrate a better effect of our method than that of the previous patch-based texture synthesis image completion algorithm.
We propose adaptive nonlinear auto-associative modeling (ANAM) based on Locally Linear Embedding algorithm (LLE) for learning intrinsic principal features of each concept separately and recognition thereby. Unlike tra...
详细信息
3D reconstruction from a single RGB image for urban scenes has been a foundation for safety-critical applications such as autonomous driving and city planning. It is essential to develop 3D reconstruction models that ...
详细信息
Constructing the pyramidal architecture for the feature is currently a very effective way to obtain feature information of objects at different scales. Although the feature pyramid can realize the recognition and dete...
详细信息
ISBN:
(数字)9781728180281
ISBN:
(纸本)9781728180298
Constructing the pyramidal architecture for the feature is currently a very effective way to obtain feature information of objects at different scales. Although the feature pyramid can realize the recognition and detection of multi-scale objects in the object detection task well, it still has some limitations. Since the feature information of different levels is often not from the same layer of the network, it is difficult to obtain the feature of different objects information at a certain scale from a certain level feature map of the pyramid network. To solve this problem, we present a novel object detection architecture, named Enhanced Multi-scale Feature Fusion Pyramid Network (EMFFPNet). Our network consists of Enhanced Multi-scale Feature Fusion Module (EMFFM) and Predictor Optimization Module (POM). In EMFFM, Features at different levels can be fused into the Enhanced features as outputs, which are more representative and deterministic. In order to enable the enhanced features to play their respective roles in the pyramid network, we assign different weights to fusion features of different levels in POM. We perform the experiments on the COCO detection benchmark. The experimental results indicate that the performance of our model is much better than the state-of-the-art model.
To solve the problem of low accuracy of gait recognition in complex scenes, a novel skeleton-based gait recognition algorithm, GCGait, is proposed. Taking human posture as the input of gait feature, the interference c...
详细信息
A fast and efficient algorithm is presented to label the connected components for binary image, especially for very huge images or any image larger than the available memory. The cascading style scheme compresses the ...
详细信息
Weak boundary contrast, inhomogeneous background and overlapped intensity distributions of the object and background are main causes that may lead to failure of boundary detection for many traditional active contour m...
详细信息
ISBN:
(纸本)9781479923427
Weak boundary contrast, inhomogeneous background and overlapped intensity distributions of the object and background are main causes that may lead to failure of boundary detection for many traditional active contour methods. In this paper, we propose a region-based active contour model to address these problems in both local and global ways. A localized active contour framework is developed, in which two local boundary measures are introduced for the evolution of the level set function. These measures are used to select the boundary candidates for boundary preservation such that the evolution of the contour is guided in a reasonable way. The object boundary is determined by a global boundary measure which evaluates the boundary completeness during the entire evolution process. The experiments demonstrate that our method works well against weak boundary contrast, inhomogeneous background and overlapped intensity distributions.
暂无评论