AIParsing: Anchor-Free Instance-Level Human Parsing.

Most state-of-the-art instance-level human parsing models adopt two-stage anchor-based detectors and, therefore, cannot avoid the heuristic anchor box design and the lack of analysis on a pixel level. To address these two issues, we have designed an instance-level human parsing network which is anchor-free and solvable on a pixel level. It consists of two simple sub-networks: an anchor-free detection head for bounding box predictions and an edge-guided parsing head for human segmentation. The anchor-free detector head inherits the pixel-like merits and effectively avoids the sensitivity of hyper-parameters as proved in object detection applications. By introducing the part-aware boundary clue, the edge-guided parsing head is capable to distinguish adjacent human parts from among each other up to 58 parts in a single human instance, even overlapping instances. Meanwhile, a refinement head integrating box-level score and part-level parsing quality is exploited to improve the quality of the parsing results. Experiments on two multiple human parsing datasets (i.e., CIHP and LV-MHP-v2.0) and one video instance-level human parsing dataset (i.e., VIP) show that our method achieves the best global-level and instance-level performance over state-of-the-art one-stage top-down alternatives.

Brain tumor segmentation based on region of interest-aided localization and segmentation U-Net.

Since magnetic resonance imaging (MRI) has superior soft tissue contrast, contouring (brain) tumor accurately by MRI images is essential in medical image processing. Segmenting tumor accurately is immensely challenging, since tumor and normal tissues are often inextricably intertwined in the brain. It is also extremely time consuming manually. Late deep learning techniques start to show reasonable success in brain tumor segmentation automatically. The purpose of this study is to develop a new region-of-interest-aided (ROI-aided) deep learning technique for automatic brain tumor MRI segmentation. The method consists of two major steps. Step one is to use a 2D network with U-Net architecture to localize the tumor ROI, which is to reduce the impact of normal tissue's disturbance. Then a 3D U-Net is performed in step 2 for tumor segmentation within identified ROI. The proposed method is validated on MICCAI BraTS 2015 Challenge with 220 high Gliomas grade (HGG) and 54 low Gliomas grade (LGG) patients' data. The Dice similarity coefficient and the Hausdorff distance between the manual tumor contour and that segmented by the proposed method are 0.876 ±0.068 and 3.594±1.347 mm, respectively. These numbers are indications that our proposed method is an effective ROI-aided deep learning strategy for brain MRI tumor segmentation, and a valid and useful tool in medical image processing.

A Hybrid Network for Large-Scale Action Recognition from RGB and Depth Modalities.

The paper presents a novel hybrid network for large-scale action recognition from multiple modalities. The network is built upon the proposed weighted dynamic images. It effectively leverages the strengths of the emerging Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) based approaches to specifically address the challenges that occur in large-scale action recognition and are not fully dealt with by the state-of-the-art methods. Specifically, the proposed hybrid network consists of a CNN based component and an RNN based component. Features extracted by the two components are fused through canonical correlation analysis and then fed to a linear Support Vector Machine (SVM) for classification. The proposed network achieved state-of-the-art results on the ChaLearn LAP IsoGD, NTU RGB+D and Multi-modal & Multi-view & Interactive ( M 2 I ) datasets and outperformed existing methods by a large margin (over 10 percentage points in some cases).

Dual-Resolution Dual-Path Convolutional Neural Networks for Fast Object Detection.

Downsampling input images is a simple trick to speed up visual object-detection algorithms, especially on robotic vision and applied mobile vision systems. However, this trick comes with a significant decline in accuracy. In this paper, dual-resolution dual-path Convolutional Neural Networks (CNNs), named DualNets, are proposed to bump up the accuracy of those detection applications. In contrast to previous methods that simply downsample the input images, DualNets explicitly take dual inputs in different resolutions and extract complementary visual features from these using dual CNN paths. The two paths in a DualNet are a backbone path and an auxiliary path that accepts larger inputs and then rapidly downsamples them to relatively small feature maps. With the help of the carefully designed auxiliary CNN paths in DualNets, auxiliary features are extracted from the larger input with controllable computation. Auxiliary features are then fused with the backbone features using a proposed progressive residual fusion strategy to enrich feature representation.This architecture, as the feature extractor, is further integrated with the Single Shot Detector (SSD) to accomplish latency-sensitive visual object-detection tasks. We evaluate the resulting detection pipeline on Pascal VOC and MS COCO benchmarks. Results show that the proposed DualNets can raise the accuracy of those CNN detection applications that are sensitive to computation payloads.

Optimal distribution of integration time for intensity measurements in degree of linear polarization polarimetry.

We consider the degree of linear polarization (DOLP) polarimetry system, which performs two intensity measurements at orthogonal polarization states to estimate DOLP. We show that if the total integration time of intensity measurements is fixed, the variance of the DOLP estimator depends on the distribution of integration time for two intensity measurements. Therefore, by optimizing the distribution of integration time, the variance of the DOLP estimator can be decreased. In this paper, we obtain the closed-form solution of the optimal distribution of integration time in an approximate way by employing Delta method and Lagrange multiplier method. According to the theoretical analyses and real-world experiments, it is shown that the variance of the DOLP estimator can be decreased for any value of DOLP. The method proposed in this paper can effectively decrease the measurement variance and thus statistically improve the measurement accuracy of the polarimetry system.

Optimal distribution of integration time for intensity measurements in Stokes polarimetry.

We consider the typical Stokes polarimetry system, which performs four intensity measurements to estimate a Stokes vector. We show that if the total integration time of intensity measurements is fixed, the variance of the Stokes vector estimator depends on the distribution of the integration time at four intensity measurements. Therefore, by optimizing the distribution of integration time, the variance of the Stokes vector estimator can be decreased. In this paper, we obtain the closed-form solution of the optimal distribution of integration time by employing Lagrange multiplier method. According to the theoretical analysis and real-world experiment, it is shown that the total variance of the Stokes vector estimator can be significantly decreased about 40% in the case discussed in this paper. The method proposed in this paper can effectively decrease the measurement variance and thus statistically improves the measurement accuracy of the polarimetric system.

Truncation Error Analysis on Reconstruction of Signal From Unsymmetrical Local Average Sampling.

The classical Shannon sampling theorem is suitable for reconstructing a band-limited signal from its sampled values taken at regular instances with equal step by using the well-known sinc function. However, due to the inertia of the measurement apparatus, it is impossible to measure the value of a signal precisely at such discrete time. In practice, only unsymmetrically local averages of signal near the regular instances can be measured and used as the inputs for a signal reconstruction method. In addition, when implemented in hardware, the traditional sinc function cannot be directly used for signal reconstruction. We propose using the Taylor expansion of sinc function to reconstruct signal sampled from unsymmetrically local averages and give the upper bound of the reconstruction error (i.e., truncation error). The convergency of the reconstruction method is also presented.

Defocus map estimation from a single image via spectrum contrast.

We present an effective method for defocus map estimation from a single natural image. It is inspired by the observation that defocusing can significantly affect the spectrum amplitude at the object edge locations in an image. By establishing the relationship between the amount of spatially varying defocus blur and spectrum contrast at edge locations, we first estimate the blur amount at these edge locations, then a full defocus map can be obtained by propagating the blur amount at edge locations over the entire image with a nonhomogeneous optimization procedure. The proposed method takes into consideration not only the affect of light refraction but also the blur texture of an image. Experimental results demonstrate that our proposed method is more reliable in defocus map estimation compared to various state-of-the-art methods.