Search

Some patients with cancer treated with programmed death 1 (PD-1) inhibitors experience immune-related severe adverse events (ir-SAEs), however, predictors are limited. The objective was to identify clinicopathologic features that may be associated with a higher ir-SAE risk. This was a nested case-control study. After screening a total of 832 PD-1 inhibitor-treated patients, we identified 42 ir-SAE cases. According to the Common Terminology Criteria for Adverse Events (CTCAE) version 5.0, ir-SAEs were defined as grade ≥3 toxic effects associated with immunotherapy. A total of 126 controls were matched. The crude and adjusted risks of ir-SAEs were estimated by odds ratio (ORs) and 95% CIs using multivariate logistic regression models. Baseline neutrophil-to-lymphocyte ratio (NLR) [per SD increment-adjusted (aOR): 1.16], lactate dehydrogenase (LDH) ≥245 U/L (aOR: 2.39), and antibiotic exposure (aOR: 4.39) were associated with a higher risk of ir-SAEs. When NLR was categorized in 3 groups, significantly higher risks of ir-SAEs (aOR: 4.95) were found in participants in group 3 (6) than in those in group 1 (3). Furthermore, NLR (per SD increment-adjusted hazard ratio:1.08) were also significantly associated with shorter overall survival (OS). Baseline LDH ≥245 U/L and antibiotic exposure were no significant association with OS. In conclusion, ir-SAEs were associated between baseline NLR, LDH ≥245 U/L and antibiotic exposure. Lower NLR was correlated with longer OS for cancer.

Few-shot segmentation aims at learning to segment query images guided by only a few annotated images from the support set. Previous methods rely on mining the feature embedding similarity across the query and the support images to achieve successful segmentation. However, these models tend to perform badly in cases where the query instances have a large variance from the support ones. To enhance model robustness against such intra-class variance, we propose a Double Recalibration Network (DRNet) with two recalibration modules, i.e., the Self-adapted Recalibration (SR) module and the Cross-attended Recalibration (CR) module. In particular, beyond learning robust feature embedding for pixel-wise comparison between support and query as in conventional methods, the DRNet further exploits semantic-aware knowledge embedded in the query image to help segment itself, which we call 'self-adapted recalibration'. More specifically, DRNet first employs guidance from the support set to roughly predict an incomplete but correct initial object region for the query image, and then reversely uses the feature embedding extracted from the incomplete object region to segment the query image. Also, we devise a CR module to refine the feature representation of the query image by propagating the underlying knowledge embedded in the support image's foreground to the query. Instead of foreground global pooling, we refine the response at each pixel in the query feature map by attending to all foreground pixels in the support feature map and taking the weighted average by their similarity; meanwhile, feature maps of the query image are also added back to weighted feature maps as a residual connection. Our DRNet can effectively address the intra-class variance under the few-shot setting with such two recalibration modules, and mine more accurate target regions for query images. We conduct extensive experiments on the popular benchmarks PASCAL- 5

Though person re-identification has witnessed significant progress, person search, as a more practical task considering the unavailability of annotations of pedestrian bounding boxes, has progressed much slower because of less discriminative feature representation. To this end, we propose a novel self-attention based person search approach by perceiving informative implicit key-points with weak supervision. Firstly, we design the Self-attention Slice Part (SSP) module, to implicitly localize informative key-points by only taking a pre-defined number of points as supervision. Concretely, this module utilizes both channel-attention and spatial attention with weak supervision on partitioned pedestrian slices to get the most discriminative key-points. After that, we strengthen the self-attention weight for these cardinal key-points, and then, more robust feature representations for conducting person search can be obtained from these self-mined key-points. Meanwhile, the SSP also provides semantic alignment with the horizontally partitioned slices. Besides, to focus more on reducing the inner-class margin rather than enlarging inter-class distance, the Random Label Smooth (RLS) loss is defined for more robust classification. The RLS loss not only provides a larger margin hyperplane but also enhances the training efficiency. Therefore, all in all: (1) We propose an end-to-end person search framework by fully exploiting current object detection and person re-identification techniques jointly with weak supervision. (2) Our proposed weakly-supervised self-attention module is generic and can be plugged into any related tasks to improve the performance. (3) We conduct extensive experiments on popular benchmarks, including the dataset of CUHK-SYSU and PRW, and our approach outperforms most current state-of-the-art methods according to mAP and top-1 evaluation metrics.

Traffic sign detection is a crucial step for automatic driving and Intelligent Transportation. Promising results have been achieved in the area of traffic sign detection, but most of them are limited to ideal environment, where the traffic signs are very clear and large. Actually, traffic sign detection is always realized based on object detection methods. However, existing object detection methods failed to detect most of the traffic signs, especially in surveillance videos or driving recorder videos. In fact, traffic signs, i.e. traffic lights, or distant road signs in driving recorded video, always cover less than 5% of the whole image in the view of camera. Therefore, in this paper, we dedicate an effort to propose a real-time small traffic sign detection approach based on revised Faster-RCNN. More specifically, firstly, we use a small region proposal generator to extract the characteristics of small traffic signs. That is to say, considering that the stride of generator is too large, we remove the pool4 layer of VGG-16 and adopt dilation for ResNet. Secondly, we combine the revised architecture of Faster-RCNN with Online Hard Examples Mining (OHEM) to make the system more robust to locate the region of small traffic signs. Finally, we conduct extensive experiments and empirical evaluations on several different videos to demonstrate the satisfying performance of our approach. i.e., the experimental results show our approach improve the mean average precision by 12.1% over the original object detection algorithm.

Crowd saliency prediction refers to predicting where people look at in crowd scene. Humans have remarkable ability to rapidly direct their gaze to select visual information of interest when looking at a visual scene. Until now, research efforts are still focused on what type of feature is representative for crowd saliency, and which type of learning model is robust for crowd saliency prediction. In this paper, we propose a Random Forest (RF) based crowd saliency prediction approach with optimal feature combination, i.e., the Feature Combination Selection for Crowd Saliency (FCSCS) framework. More specifically, we first define three representative crowd saliency features, namely, FaceSizeDiff, FacePoseDiff and FaceWhrDiff. Next, we adopt the Random Forest (RF) algorithm to construct our saliency learning model. Then, we evaluate the performance of FCSCS framework with different feature combinations (fifteen combinations in our experiments). Those selected features include low-level features (i.e., color, intensity, orientation), four crowd features (i.e., face size, face density, frontal face, profile face) and three new defined features (i.e., FaceSizeDiff, FacePoseDiff and FaceWhrDiff). We use FCSCS framework to obtain the optimal feature combination that is most suitable for crowd saliency prediction and further train the saliency model based on the optimal feature combination. After that, we evaluate the performance of the crowd saliency prediction classifiers. Finally, we conduct extensive experiments and empirical evaluation to demonstrate the satisfactory performance of our approach.