Hydrogel scaffolds are attractive for tissue defect repair and reorganization because of their human tissue-like characteristics. However, most hydrogels offer limited cell growth formation ability due to submicron- or nano-sized gel networks, which restrict the supply oxygen, nutrients inhibit proliferation differentiation encapsulated cells. In recent years, 3D printed have shown great potential overcome this problem by introducing macro-pores within scaffolds. study, we fabricated a...

Abstract Hydrogel coating has a certain ability to inhibit marine biofouling. However, owing their poor mechanical properties and short service life, they have not found solid applications in protection. Herein, novel self‐regenerating zwitterionic hydrogel composed of self‐crosslinking degradable polymer with hydrolysis‐induced zwitterions corrosion inhibitor benzotriazole (BTA) loaded metal‐organic framework (MOF) is reported, which exhibits both anti‐biofouling anti‐corrosion performance....

Semantic segmentation and height estimation are two critical tasks in remote sensing scene understanding that highly correlated with each other. To address both simultaneously, it is natural to consider designing a unified deep learning model aims improve performance by jointly complementary information among the associated tasks. In this paper, we learn under multi-task framework propose novel objective functions, called cross-task contrastive loss cross-pixel loss, respectively, enhance...

In recent years, object detectors generally use the feature pyramid network (FPN) to solve problem of scale variation in detection. this paper, we propose a new architecture which combines top-down and bottom-up network. The main contributions proposed method are two-fold: (1) We design more complex get maps for (2) By combining these two architectures, can with richer semantic information better. experiments on PASCAL VOC2007 dataset. Experimental results show that improve accuracy using...

Tough hydrogels with the ability to be repeatedly processed into various shapes as thermoplastics are highly desired in advanced medical devices and tissue engineering. Here, we have developed a kind of versatile supramolecular hydrogel network cross-linked by double hydrogen bonds from poly(N-acryloyl glycinamide) (PNAGA). The resulting PNAGA-30 (30 wt% solid content) tough, re-processable, recyclable similar thermoplastics. form fragments can easily re-processed including sheet, filament,...

The low back pain (LBP) caused by intervertebral disc (IVD) herniation and degeneration bring huge physical economic burden to individuals families. Since current clinical treatment methods could not change the process of disc, annulus fibrosus (AF) repair as a method seal defects IVD prevent reherniation after lumbar discectomy has attracted increasing interest. In this study, porous HA-PEG/NAGA-GelMA double network hydrogel with multiple hydrogen bonds was designed prepared two-step...

The challenges of learning a robust 6D pose function lie in 1) severe occlusion and 2) systematic noises depth images. Inspired by the success point-pair features, goal this paper is to recover an object instance segmented from RGB-D images locally matching pairs oriented points between model camera space. To end, we propose novel Bi-directional Correspondence Mapping Network (BiCo-Net) first generate point clouds guided typical regression, which can thus incorporate pose-sensitive...

In previous studies, the predictive role of BIM deletion polymorphism with respect to responses epidermal growth factor receptor tyrosine kinase inhibitors (EGFR-TKIs) has been controversial. The potential reasons for these inconsistent findings were unknown.

Height estimation from a single remote sensing image has great potential in generating digital surface models (DSM) efficiently for quick earth reconstruction. Recently, convolutional neural networks (CNN) have emerged as powerful method to deal with this ill-posed problem. Most existing methods formulate height regression problem due the continuity of object height. However, it is difficult model regress heights exactly ground-truth values wide range. In letter, we reformulate task...

We propose a wav2vec based speech emotion detection framework, combining features from pretrained models with neural or traditional classifiers and achieves the best performance compared other baselines, conduct result analysis.

Abstract Air pollution remains to be a challenge for the government and scholars, especially particulate matter (PM 2.5 ), resulting in many casualties huge economic losses. It is imperative understand distribution characteristics of PM analysis some influencing factors including wind speed, humidity atmospheric boundary conditions. A certain scholars have studied relationship between condition. However, there no study about correlation speed or humidity. Considering all components...

Data augmentation has always been an effective way to overcome overfitting issue when the dataset is small. There are already lots of operations such as horizontal flip, random crop or even Mixup. However, unlike image classification task, we cannot simply perform these for object detection task because lack labeled bounding boxes information corresponding generated images. To address this challenge, propose a framework making use Generative Adversarial Networks(GAN) unsupervised data...

Previous approaches to the problem of generalization for out-of-distribution (OOD) data usually assume that from each environment is available simultaneously, which unrealistic in real-world applications. In this paper, we develop a new framework termed sequential invariant information bottleneck (seq-IIB) improve ability learning agents environments. Our main idea combine merits famed Information Bottleneck (IB) principle with Invariant Risk Minimization (IRM), such agent can gradually...

Domain gap between synthetic and real data in visual regression (e.g. 6D pose estimation) is bridged this paper via global feature alignment local refinement on the coarse classification of discretized anchor classes target space, which imposes a piece-wise manifold regularization into domain-invariant representation learning. Specifically, our method incorporates an explicit self-supervised regularization, revealing consistent cumulative dependency across domains, to self-training scheme...

Domain gap between synthetic and real data in visual regression (e.g., 6D pose estimation) is bridged this paper via global feature alignment local refinement on the coarse classification of discretized anchor classes target space, which imposes a piece-wise manifold regularization into domain-invariant representation learning. Specifically, our method incorporates an explicit self-supervised regularization, revealing consistent cumulative dependency across domains, to self-training scheme...

This paper presents a proposal to detect the High-Latitude Dust (HLD) events on images captured by NASA satellites. We plan use semantic segmentation techniques [1] identify HLD regions at pixel level. The proposed model would distribute labels specific pixels that we need identify. After analyzing and learning those attributes, trained in new with events. However, training, imbalanced data leads gap model's performance coastal areas sea areas, which means even though of dust detection over...

Is it possible to develop an "AI Pathologist" pass the board-certified examination of American Board Pathology (ABP)? To build such a system, three challenges need be addressed. First, we create visual question answering (VQA) dataset where AI agent is presented with pathology image together and asked give correct answer. Due privacy concerns, images are usually not publicly available. Besides, only well-trained pathologists can understand images, but they barely have time help datasets for...

Contrastive self-supervised learning has attracted significant research attention recently. It learns effective visual representations from unlabeled data by embedding augmented views of the same image close to each other while pushing away embeddings different images. Despite its great success on ImageNet classification, COCO object detection, etc., performance degrades contrast-agnostic applications, e.g., medical where all images are visually similar other. This creates difficulties in...