Search

6D object pose estimation for texture-less objects from RGB images remains challenging, especially in occlusion scenarios. Instead of localizing sparse keypoints by regressing their image coordinates or heatmaps, which are sensitive to occlusion, we introduce GeoPose, a novel reconstruction guided pose estimation pipeline that predicts dense correspondences and leverages geometric consistency effectively. We first design a dense reconstruction network (ReconNet) to reconstruct pixel-wise object coordinates in normalization space. Dense 2D-3D correspondences are generated intuitively by our explicit pa-rameterization for 3D object models, which dismisses keypoint selection efforts. These 2D-3D correspondences are then utilized to estimate 6D poses by the PnP algorithm with RANSAC iterations. Furthermore, a novel Cycle Loss is proposed to provide 3D prior supervision, which significantly correlates with the pose estimation task by guiding geometric consistency between reconstruction (pixel to 3D) and reprojection (3D to pixel). In addition, a training data augmentation method is proposed to handle the insufficiency of 6D datasets, the acquisition of which is error-prone and time-consuming. Extensive experiments demonstrate that, compared with existing RGB-based methods, our GeoPose can achieve state-of-the-art (SOTA) 6D pose estimation performance on the LINEMOD, Occlusion LINEMOD and T-LESS datasets.

This paper aims to solve the problem of estimating the 6D pose of an object under occlusion using RGB-D images. Most existing methods typically use the information of color and depth images separately to make predictions, which limits their performances in the presence of occlusion. Instead, we propose a pipeline to effectively fuse color and depth information and perform region-level pose estimation. Our method first uses a CNN to extract the color features, and then we obtain the fusion features by combining the color features into the point cloud. Unlike existing methods, the fusion features are in the form of point sets instead of feature maps. We further use a PointNet++-like network to process the fusion features, obtaining several region-level features. Each region-level feature can predict a pose with confidence. The pose with the highest confidence is chosen as the final output. Experiments show that the proposed method outperforms the state-of-the-art methods on both the LINEMOD and Occlusion LINEMOD datasets, indicating that the proposed pipeline can obtain accurate pose estimation results and is robust to occlusion.

In product inspection scenarios, the objects have different characteristics of textures and different potential anomalies. Still, traditional methods cannot precisely segment structural anomalies and achieve optimal performance on all object categories. This article introduces a flexible multilevel image reconstruction (MLIR) framework for the unsupervised anomaly segmentation task. The MLIR integrates multiple generators with different downsampling ratios to serve as multiple reconstruction levels and leverage an adjustable perceptual similarity measurement to obtain the featurewise reconstruction errors, overcoming the challenge of detecting multiscale anomalies. Leveraging embeddings with different receptive fields for reconstruction, generators perform differently on the same target. Therefore, we depend on the reconstruction uncertainty to propose an adaptive attention-level transition (ALT) strategy that optimizes the detection performance automatically according to the target’s category. ALT adjusts the weights of both reconstruction levels and feature measurement scales simultaneously to utilize features at a consistent level for reconstruction and anomaly detection (AD). This strategy remarkably improves the system’s adaptivity for different applications. Moreover, we propose reducing the signal-to-noise ratio (SNR) of the images by adding pepper noise so that the reconstruction process is framed as an image denoising task, which improves feature extraction and image restoration robustness. We conduct extensive experiments on the MVTec AD dataset and demonstrate that the proposed method performs satisfactorily to different categories and achieves the state-of-the-art (SOTA) performance.

Background Piglet mortality is an economically important complex trait that impacts sow prolificacy in the pig industry. The genetic parameters estimations and genome-wide association studies will help us to better understand the genetic fundamentals of piglet mortality. However, compared with other economically important traits, a little breakthrough in the genetic analyses of the trait has been achieved. Results In this study, we used multi-breed data sets from Yorkshire, Landrace, and Duroc sows and characterized the genetic and genomic properties of mortality rate at birth by treating each parity as a different trait. The heritability of mortality rate from parity I to III were estimated to be 0.0630, 0.1031, and 0.1140, respectively. The phenotypic and genetic correlations with its component traits were all positive with ranges from 0.0897 to 0.9054, and 0.2388 to 0.9999, respectively. Integrating the results, we identified 21 loci that were detected at least by two tools from standard MLM, FarmCPU, BLINK and mrMLM, and these loci were annotated to 22 genes. The annotations revealed that the gene expressions were associated with the reproductive system, nervous system, digestive system, and embryonic development, which are reasonably related to the piglet mortality. Conclusions In brief, the genetic properties of piglet mortality at birth were reported. These findings are expected to provide much information for understanding the genetic and genomic fundamentals of farrowing mortality and also identify candidate molecular markers for breeding practice.

In this paper, we study the performance of different deep convolutional networks in 1:N face recognition and process the feature comparison in order to obtain a higher face recognition rate. The improved feature comparison method is proposed: using Basis Pursuit De-Noising (BPDN) method combined with Homotopy algorithm to replace the nearest neighbor analysis in 1:N face recognition. Based on the Facenet model, we also compare the performance of different loss functions in face verification and recognition. Furthermore, the ensemble method and voting classifier are used to integrate Facenet, InsightFace and BPDN method to realize 1:N face recognition. It should be noted that the 1:N dataset is transformed from Labeled Faces in the Wild (LFW) dataset, including the 1 dataset and the N dataset. The 1 dataset and the N dataset contain 1680 faces and 5749 faces respectively, one for each person. The final experiments show that the ensemble method's accuracy is 82.98%, which is higher than the above methods, and indicate the ensemble method we proposed can improve the face recognition rate.

Vehicle face detection is an important and challenging task in face detection. On one hand, the face image is affected by the complex environment during driving so that the facial details are easily lost. On the other hand, there is no large public vehicle face dataset. In order to solve these problems, firstly, inspired by the convolutional neural network, an improved method based on cascaded convolutional neural networks is proposed in this paper. The cascaded convolutional networks consist of three lightened convolutional neural networks. Each convolutional neural network uses multi-task learning. Feature fusion technology focusing on learning details of vehicle face image is also adopted in cascaded convolutional networks. Secondly, due to the lack of vehicle face image samples, we build a small vehicle face dataset by ourselves, named Driver FACE dataset. Experiments on Driver FACE dataset show that our method improves the performance of vehicle face detection compared with baseline.

In spired by the fact that a logarithm difference is the subtraction of two neighbor pixels, which may be a positive or negative numerical value, we divide a face local region into a positive region and a negative region, and the general logarithm difference model (GLDM) is developed by integrating positive and negative logarithm differences. Then, the multiscale logarithm difference edge-maps (MSLDE) [1] is employed as the test-bed, and the proposed GLDM is introduced into MSLDE to form General MSLDE (GMSLDE). Finally, the performance of GMSLDE is verified on the Extended Yale B and CMU PIE face databases with severe illumination variations. The experimental results indicate that the proposed GLDM can efficiently improve the performance of logarithm difference edge-map against severe illumination variations.

In recent years, with the advent of large data sets and the adoption of end-to-end learning methods using convolutional networks, it is possible to verify and recognize faces on a large-scale dataset effectively, which has even surpassed that of humans. One of the most effective methods is the metric learning with triplet loss. However, when using a large-scale dataset to train the model, the number of the possible triplets will be explosive. Therefore, we propose an improved triplet loss based on deep neural network for end-to-end metric learning, which can effectively cut down the number of the possible triplets. It is a simplification of the traditional triplet loss function, but it has not degraded the performance of the model. Our method is evaluated on LFW and YTF datasets, which are most widely used benchmarks. The experimental results indicate that the proposed method can outperform some previous methods, and obtain the state-of-art performance with less time.

To study the influence of estradiol on histomorphology of skin flaps with ischemia reperfusion injury.48 adult male Wistar rats aged 12-14 weeks old, were randomly divided into control group (group I), ischemia-reperfusion group (group II), saline group (group III), estradiol group (group IV). Superficial epigastric artery axial flap, 3 cm x 6 cm in size, was made in the left lower quadrant abdominal of each rat. Flap model with ischemia-reperfusion injury was established by using the nondestructive micro vascular clamp to clamp the superficial epigastric artery. The general condition of the flap was observed after operation. At 7 days after operation, the survival rate of the flap was detected, the flaps were harvested to receive histology and ultrastructural observation. The neutrophils level of the superficial epigastric vein were tested.7 days after operation, the survival rate of the flap in group IV was significantly higher than that in group II, III (P0.05). The neutrophils level in group IV was lower than that in group II, III (P0.05). The histological observation showed that the degree of tissue swelling and inflammatory exudation in group IV was more slight than that in group II, III. Presence of high neutrophils density were observed in group II, III, while slight inflammation and necrosis were observed in group IV. In group I, collagen fibers in flap are regularly arranged with no significant necrosis. Oganelles structure disappeared and apoptotic bodies were shown in group II and group III, even the lysosome could be seen in the cell. Collagen fibers in flap are regularly arranged with slight swelling and no obvious ultrastructural necrocytosis was seen in the cell of group IV.The estradiol can significantly increase flap survival rate by inhibiting neutrophils infiltration and improving the pathological changes of organization structure in flap.

This paper determines two solutions of Emergency resource distribution, according to the scene of disaster and the supplier distribution of emergency resource, the types and quantities of the inventory resource. The first one is the single save point -- multi-disaster points integer programming model which offers three cases to determine the optimization distribution solution, and uses optimization software such as Lindo or Lingo to solve the types and quantities of resource distribution. Another one is multi-depot and single disaster points model, which means that different emergency resource suppliers allocate resources to more complex disaster areas, and Tops is method is adopted to solve it. Moreover, two examples demonstrate the applicability of these two models.