Cell-free DNA methylation reveals cell-specific tissue injury and correlates with disease severity and patient outcomes in COVID-19.

BACKGROUND: The recently identified methylation patterns specific to cell type allows the tracing of cell death dynamics at the cellular level in health and diseases. This study used COVID-19 as a disease model to investigate the efficacy of cell-specific cell-free DNA (cfDNA) methylation markers in reflecting or predicting disease severity or outcome. METHODS: Whole genome methylation sequencing of cfDNA was performed for 20 healthy individuals, 20 cases with non-hospitalized COVID-19 and 12 cases with severe COVID-19 admitted to intensive care unit (ICU). Differentially methylated regions (DMRs) and gene ontology pathway enrichment analyses were performed to explore the locus-specific methylation difference between cohorts. The proportion of cfDNA derived from lung and immune cells to a given sample (i.e. tissue fraction) at cell-type resolution was estimated using a novel algorithm, which reflects lung injuries and immune response in COVID-19 patients and was further used to evaluate clinical severity and patient outcome. RESULTS: COVID­19 patients had globally reduced cfDNA methylation level compared with healthy controls. Compared with non-hospitalized COVID-19 patients, the cfDNA methylation pattern was significantly altered in severe patients with the identification of 11,156 DMRs, which were mainly enriched in pathways related to immune response. Markedly elevated levels of cfDNA derived from lung and more specifically alveolar epithelial cells, bronchial epithelial cells, and lung endothelial cells were observed in COVID-19 patients compared with healthy controls. Compared with non-hospitalized patients or healthy controls, severe COVID-19 had significantly higher cfDNA derived from B cells, T cells and granulocytes and lower cfDNA from natural killer cells. Moreover, cfDNA derived from alveolar epithelial cells had the optimal performance to differentiate COVID-19 with different severities, lung injury levels, SOFA scores and in-hospital deaths, with the area under the receiver operating characteristic curve of 0.958, 0.941, 0.919 and 0.955, respectively. CONCLUSION: Severe COVID-19 has a distinct cfDNA methylation signature compared with non-hospitalized COVID-19 and healthy controls. Cell type-specific cfDNA methylation signature enables the tracing of COVID-19 related cell deaths in lung and immune cells at cell-type resolution, which is correlated with clinical severities and outcomes, and has extensive application prospects to evaluate tissue injuries in diseases with multi-organ dysfunction.

A novel incomplete hesitant fuzzy information supplement and clustering method for large-scale group decision-making.

Clustering is an effective means to reduce the scaling of large-scale group decision-making (LSGDM). However, there are many problems with clustering methods, such as incomplete or ambiguous information usually provided by different decision makers. Traditional clustering methods may not be able to handle these situations effectively, resulting in incomplete decision-making information. Calculating the clustering centers may become very complex and time-consuming. Inappropriate distance weights may also lead to incorrect cluster assignments, and these problems will seriously affect the clustering results. This research provides a novel incomplete hesitant fuzzy information supplement and clustering approach for large-scale group decision-making in order to address the aforementioned difficulties. First, the approach takes into account the trust degradation and the inhibition of relationships of distrust in the process of trust propagation, and then it builds a global and local network of trust. A novel supplemental formula is provided that takes into account the decision-preference maker's as well as the trust-neighbor's information, allowing the decision-neighbor maker's recommendation to be realized. Therefore, an improved distance function can be proposed to calculate the weights by combining the relative standard deviation theory and selecting the selected clustering centers by using the density peaks in order to optimize the selection of clustering centers and reduce the complexity and scaling of the decision. Finally, an example is presented to demonstrate how the proposed method can be applied. The consistency index and comparison experiments are used to evaluate if the suggested approach is effective and reliable.

Reprimo (RPRM) mediates neuronal ferroptosis via CREB-Nrf2/SCD1 pathways in radiation-induced brain injury.

Neuronal ferroptosis has been found to contribute to degenerative brain disorders and traumatic and hemorrhagic brain injury, but whether radiation-induced brain injury (RIBI), a critical deleterious effect of cranial radiation therapy for primary and metastatic brain tumors, involves neuronal ferroptosis remains unclear. We have recently discovered that deletion of reprimo (RPRM), a tumor suppressor gene, ameliorates RIBI, in which its protective effect on neurons is one of the underlying mechanisms. In this study, we found that whole brain irradiation (WBI) induced ferroptosis in mouse brain, manifesting as alterations in mitochondrial morphology, iron accumulation, lipid peroxidation and a dramatic reduction in glutathione peroxidase 4 (GPX4) level. Moreover, the hippocampal ferroptosis induced by ionizing irradiation (IR) mainly happened in neurons. Intriguingly, RPRM deletion protected the brain and primary neurons against IR-induced ferroptosis. Mechanistically, RPRM deletion prevented iron accumulation by reversing the significant increase in the expression of iron storage protein ferritin heavy chain (Fth), ferritin light chain (Ftl) and iron importer transferrin receptor 1 (Tfr1), as well as enhancing the expression of iron exporter ferroportin (Fpn) after IR. RPRM deletion also inhibited lipid peroxidation by abolishing the reduction of GPX4 and stearoyl coenzyme A desaturase-1 (SCD1) induced by IR. Importantly, RPRM deletion restored or even increased the expression of nuclear factor, erythroid 2 like 2 (Nrf2) in irradiated neurons. On top of that, compromised cyclic AMP response element (CRE)-binding protein (CREB) signaling was found to be responsible for the down-regulation of Nrf2 and SCD1 after irradiation, specifically, RPRM bound to CREB and promoted its degradation after IR, leading to a reduction of CREB protein level, which in turn down-regulated Nrf2 and SCD1. Thus, RPRM deletion recovered Nrf2 and SCD1 through its impact on CREB. Taken together, neuronal ferroptosis is involved in RIBI, RPRM deletion prevents IR-induced neuronal ferroptosis through restoring CREB-Nrf2/SCD1 pathways.

Loss of RSR1 function increases the abscisic acid content and improves rice quality performance at high temperature.

Rice starch regulator1 (RSR1) participates in the regulation of starch synthesis in rice, but it's function on starch synthesis and quality formation in response to high temperature is unknown. RSR1 mutation resulted in a significant increase in the abscisic acid (ABA) content in rice grains under both normal and high temperature, and the effect of high temperature on grain filling and quality formation of the rsr1 mutants was significantly reduced. The grain size, 1000-kernels weight, amylose content, gelatinization temperature, and starch viscosity of the rsr1 mutants were less sensitive to high temperature. Loss of RSR1 function increased the expression levels of starch synthesis-related genes and reduced their responses to high temperature to some extent. Besides, the percentage of germinated seeds from rsr1 mutants was significantly lower than that of the wild-type, and the difference was more significant under ABA treatment. The shoot lengths of the rsr1 mutants were remarkably shorter than those of the wild-type, which was further exacerbated by ABA treatment. These results indicated that loss function of RSR1 can improve rice quality performance at high temperature by moderately increasing the ABA content of rice grains, which provides theoretical significance for the cultivation of better-quality rice with high-temperature resistance.

Reprimo (RPRM) as a Potential Preventive and Therapeutic Target for Radiation-Induced Brain Injury via Multiple Mechanisms.

Patients receiving cranial radiotherapy for primary and metastatic brain tumors may experience radiation-induced brain injury (RIBI). Thus far, there has been a lack of effective preventive and therapeutic strategies for RIBI. Due to its complicated underlying pathogenic mechanisms, it is rather difficult to develop a single approach to target them simultaneously. We have recently reported that Reprimo (RPRM), a tumor suppressor gene, is a critical player in DNA damage repair, and RPRM deletion significantly confers radioresistance to mice. Herein, by using an RPRM knockout (KO) mouse model established in our laboratory, we found that RPRM deletion alleviated RIBI in mice via targeting its multiple underlying mechanisms. Specifically, RPRM knockout significantly reduced hippocampal DNA damage and apoptosis shortly after mice were exposed to whole-brain irradiation (WBI). For the late-delayed effect of WBI, RPRM knockout obviously ameliorated a radiation-induced decline in neurocognitive function and dramatically diminished WBI-induced neurogenesis inhibition. Moreover, RPRM KO mice exhibited a significantly lower level of acute and chronic inflammation response and microglial activation than wild-type (WT) mice post-WBI. Finally, we uncovered that RPRM knockout not only protected microglia against radiation-induced damage, thus preventing microglial activation, but also protected neurons and decreased the induction of CCL2 in neurons after irradiation, in turn attenuating the activation of microglial cells nearby through paracrine CCL2. Taken together, our results indicate that RPRM plays a crucial role in the occurrence of RIBI, suggesting that RPRM may serve as a novel potential target for the prevention and treatment of RIBI.

Effects of Music, Massage, Exercise, or Acupuncture in the Treatment of Depression Among College Students: A Network Meta-Analysis.

Objective: To assess the therapeutic impacts of exercise, massage, and music interventions on college students experiencing depression by employing a mesh meta-analysis approach. This research intends to offer valuable insights to aid in the development of non-pharmaceutical treatment strategies for depression. Methods: We conducted a thorough search across various databases including Cochrane, PubMed, Embase, Web of Science, CNKI, and Wanfang to explore the effects of music, massage, aerobic exercise, fitness Qigong, yoga, tai chi, ball games, strength training, dance, whole body vibration training, and high-intensity interval training on the treatment of depression in college students. The search period was from January 1, 2023, which marks the establishment of each database. Subsequently, a mesh meta-analysis was performed using the "Stata 15.1" software, incorporating outcome indicators from 24 included literature comprising a total of 1458 patients. Results: Based on the ranking of the optimal intervention effects of various non-pharmaceutical methods, the order, from highest to lowest probability, was as follows: high-intensity interval training (96%), yoga (94.90%), dance (78.30%), music (73.30%), ball games (62.50%), strength training (51.70%), aerobic training (45.30%), tai chi (35.40%), vibration training (27.30%), massage (20.10%), qigong (14.30%), and no intervention (1.00%). This ranking aligns closely with the findings obtained from pairwise comparisons between different interventions. Conclusion: High-intensity interval training is likely to yield the most effective therapeutic results for college students with depression. In the pairwise comparison of different interventions, High-intensity interval training is also better than most interventions. However, to establish its intervention effect more conclusively, further validation through additional high-quality randomized controlled trials is necessary.

Protocol for IPO11 deletion and re-expression in H460 lung cancer cells using CRISPR-Cas9 and plasmid transfection.

Exploring the essential role of Importin 11 (IPO11) in the nuclear translocation of its potential cargo proteins requires an efficient means of IPO11 deletion and re-expression. Here, we present a protocol for the generation of IPO11 deletion using CRISPR-Cas9 and re-expression using plasmid transfection in H460 non-small cell lung cancer cells. We describe steps for lentiviral transduction of H460 cells, single clone selection, and expansion and validation of cell colonies. We then detail plasmid transfection and validation of transfection efficiency. For complete details on the use and execution of this protocol, please refer to Zhang et al.1.

Structured Knowledge Distillation for Dense Prediction.

In this work, we consider transferring the structure information from large networks to compact ones for dense prediction tasks in computer vision. Previous knowledge distillation strategies used for dense prediction tasks often directly borrow the distillation scheme for image classification and perform knowledge distillation for each pixel separately, leading to sub-optimal performance. Here we propose to distill structured knowledge from large networks to compact networks, taking into account the fact that dense prediction is a structured prediction problem. Specifically, we study two structured distillation schemes: i) pair-wise distillation that distills the pair-wise similarities by building a static graph; and ii) holistic distillation that uses adversarial training to distill holistic knowledge. The effectiveness of our knowledge distillation approaches is demonstrated by experiments on three dense prediction tasks: semantic segmentation, depth estimation and object detection. Code is available at https://git.io/StructKD.

Short-term outcomes of near-infrared imaging using indocyanine green in laparoscopic lateral pelvic lymph node dissection for middle-lower rectal cancer: A propensity score-matched cohort analysis.

Laparoscopic lateral pelvic lymph node dissection (LPND) is limited by complex neurovascular bundles in the narrow pelvic sidewall and various post-operative complications. Indocyanine green (ICG) has been applied to increase the number of harvested lymph nodes and reduce the injury of irrelevant vessels in patients with rectal cancer. However, few studies on the recurrence rate of ICG fluorescence imaging-guided laparoscopic LPND were reported. This retrospective study enrolled 50 middle- low rectal cancer patients who were treated by LPND. After propensity score matching, 20 patients were matched in each of the indocyanine green (ICG) guided imaging group (ICG group) and non-ICG guided imaging group (non-ICG group). The average follow-up time was 13.5 months (12-15 months). Our results showed that the total number of harvested lymph nodes in the ICG group was significantly higher than that in the non-ICG group (P < 0.05), and intraoperative blood loss and post-operative hospital stay times in the ICG group were less than those in the non-ICG group (P < 0.05). After 12 months of follow-up, no residual lymph node and local tumor recurrence were found for patients in the ICG group. Four patients in the non-ICG group detected residual lymph nodes at the 3-month visit. Our findings highlighted the importance of ICG fluorescence-guided imaging in LPND because it has unique advantages in improving the number of lymph node dissections, surgical accuracy, and decreasing the residual lymph nodes and local tumor recurrence. In addition, ICG fluorescence guidance technology can effectively shorten the operation time, and it is simple to operate, which is worth popularizing.

RPRM negatively regulates ATM levels through its nuclear translocation on irradiation mediated by CDK4/6 and IPO11.

How the ataxia telangiectasia mutated (ATM) protein kinase, a core protein in DNA damage response, is regulated at post-transcription level remains unclear. Here it is identified that protein Reprimo (RPRM) downregulates ATM protein levels, resulting in impaired DNA repair and enhanced cellular radiosensitivity. Mechanistically, although primarily localized in the cytoplasm, RPRM translocates to the nucleus shortly after induced by X-irradiation, interacts with ATM and promotes its nuclear export and proteasomal degradation. The RPRM nuclear translocation involves its phosphorylation at serine 98 mediated by cyclin-dependent kinases 4/6 (CDK4/6), and requires Importin-11 (IPO11). Of importance, IPO11-regulated RPRM nuclear import upon irradiation is essential for its regulation on ATM. Thus, RPRM overexpression and its phosphorylation inhibition sensitize cells to genotoxic agents such as irradiation, whereas RPRM deficiency significantly increases resistance to radiation-induced damage both in vitro and in vivo. These findings establish a crucial regulatory mechanism in which ATM is negatively modulated by RPRM.

Formulaic language identification model based on GCN fusing associated information.

Formulaic language is a general term for ready-made structures in a language. It usually has fixed grammatical structure, stable language expression meaning and specific use context. The use of formulaic language can coordinate sentence generation in the process of writing and communication, and can significantly improve the idiomaticity and logic of machine translation, intelligent question answering and so on. New formulaic language is generated almost every day, and how to accurately identify them is a topic worthy of research. To this end, this article proposes a formulaic language identification model based on GCN fusing associated information. The innovation is that each sentence is constructed into a graph in which the nodes are part-of-speech features and semantic features of the words in the sentence and the edges between nodes are constructed according to mutual information and dependency syntactic relation. On this basis, the graph convolutional neural network is adopted to extract the associated information between words to mine deeper grammatical features. Therefore, it can improve the accuracy of formulaic language identification. The experimental results show that the model in this article is superior to the classical formulaic language identification model in terms of accuracy, recall and F1-score. It lays a foundation for the follow-up research of formulaic language identification tasks.

UBE2T regulates FANCI monoubiquitination to promote NSCLC progression by activating EMT.

Fanconi anemia complementation group I (FANCI) is a critical protein for maintaining DNA stability. However, the exact role of FANCI in tumors remains to be elucidated. The present study aimed to explore the role and potential mechanism of action of FANCI in non­small cell lung cancer (NSCLC). To quantify the expression levels of FANCI and ubiquitin­conjugating enzyme E2T (UBE2T) in NSCLC tissues, reverse­transcription quantitative PCR and western blotting were employed. Cell Counting Kit­8, wound healing and Transwell assays along with flow cytometry analysis and tumor xenograft were used to investigate the biological effects of FANCI in NSCLC in vitro and in vivo. The binding of FANCI with UBE2T was confirmed using a co­immunoprecipitation assay. Epithelial­to­mesenchymal transition (EMT) protein markers were quantified via western blotting. The results showed that FANCI expression level was higher in NSCLC tumor tissues, compared with adjacent tissues. In A549 and H1299 cells, knockdown of FANCI inhibited cell proliferation, migration, invasion, cell cycle and EMT in vitro. Tumor growth was repressed in vitro, upon downregulation of FANCI expression. UBE2T was observed to directly bind to FANCI and regulate its monoubiquitination. Overexpression of UBE2T reversed the effects induced by FANCI knockdown in NSCLC cells. Furthermore, it was noted that FANCI interacted with WD repeat domain 48 (WDR48). Overexpression of WDR48 reversed the effects of FANCI on cell proliferation, migration and EMT. In conclusion, FANCI was identified to be a putative oncogene in NSCLC, wherein FANCI was monouniubiquitinated by UBE2T to regulate cell growth, migration and EMT through WDR48. The findings suggested that FANCI could be used as a prognostic biomarker and therapeutic target for NSCLC.

Irradiated microvascular endothelial cells may induce bystander effects in neural stem cells leading to neurogenesis inhibition.

Radiation-induced neurocognitive dysfunction (RIND) has attracted a lot of attention lately due to the significant improvement of the survival of cancer patients after receiving cranial radiotherapy. The detailed mechanisms are not completely understood, but extensive evidence supports an involvement of the inhibition of hippocampal neurogenesis, which may result from radiation-induced depletion of neural stem cells (NSCs) as well as the damage to neurogenic niches. As an important component of neurogenic niches, vascular cells interact with NSCs through different signaling mechanisms, which is similar to the characteristics of radiation-induced bystander effect (RIBE). But whether RIBE is involved in neurogenesis inhibition contributed by the damaged vascular cells is unknown. Thus, the purpose of the present study was to investigate the occurrence of RIBEs in non-irradiated bystander NSCs induced by irradiated bEnd.3 vascular endothelial cells in a co-culture system. The results show that compared with the NSCs cultured alone, the properties of NSCs were significantly affected after co-culture with bEnd.3 cells, and further change was induced without obvious oxidative stress and apoptosis when bEnd.3 cells were irradiated, manifesting as a reduction in the proliferation, neurosphere-forming capability and differentiation potential of NSCs. All these results suggest that the damaged vascular endothelial cells may contribute to neurogenesis inhibition via inducing RIBEs in NSCs, thus leading to RIND.

Distillation-Guided Residual Learning for Binary Convolutional Neural Networks.

It is challenging to bridge the performance gap between binary convolutional neural network (BCNN) and floating-point CNN (FCNN). This performance gap is mainly caused by the inferior modeling capability and training strategy of BCNN, which leads to substantial residuals in intermediate feature maps between BCNN and FCNN. To minimize the performance gap, we enforce BCNN to produce similar intermediate feature maps with the ones of FCNN. This intuition leads to a more effective training strategy for BCNN, i.e., optimizing each binary convolutional block with blockwise distillation loss derived from FCNN. The goal of minimizing the residuals in intermediate feature maps also motivates us to update the binary convolutional block architecture to facilitate the optimization of blockwise distillation loss. Specifically, a lightweight shortcut branch is inserted into each binary convolutional block to complement residuals at each block. Benefited from its squeeze-and-interaction (SI) structure, this shortcut branch introduces a fraction of parameters, e.g., less than 10% overheads, but effectively boosts the modeling capability of binary convolution blocks in BCNN. Extensive experiments on ImageNet demonstrate the superior performance of our method in both classification efficiency and accuracy, e.g., BCNN trained with our methods achieves the accuracy of 60.45% on ImageNet, better than many state-of-the-art ones.

Few-Shot Image and Sentence Matching via Aligned Cross-Modal Memory.

Image and sentence matching has attracted much attention recently, and many effective methods have been proposed to deal with it. But even the current state-of-the-arts still cannot well associate those challenging pairs of images and sentences containing few-shot content in their regions and words. In fact, such a few-shot matching problem is seldom studied and has become a bottleneck for further performance improvement in real-world applications. In this work, we formulate this challenging problem as few-shot image and sentence matching, and accordingly propose an Aligned Cross-Modal Memory (ACMM) model to deal with it. The model can not only softly align few-shot regions and words in a weakly-supervised manner, but also persistently store and update cross-modal prototypical representations of few-shot classes as references, without using any groundtruth region-word correspondence. The model can also adaptively balance the relative importance between few-shot and common content in the image and sentence, which leads to better measurement of overall similarity. We perform extensive experiments in terms of both few-shot and conventional image and sentence matching, and demonstrate the effectiveness of the proposed model by achieving the state-of-the-art results on two public benchmark datasets.

Whole brain segmentation with full volume neural network.

Whole brain segmentation is an important neuroimaging task that segments the whole brain volume into anatomically labeled regions-of-interest. Convolutional neural networks have demonstrated good performance in this task. Existing solutions, usually segment the brain image by classifying the voxels, or labeling the slices or the sub-volumes separately. Their representation learning is based on parts of the whole volume whereas their labeling result is produced by aggregation of partial segmentation. Learning and inference with incomplete information could lead to sub-optimal final segmentation result. To address these issues, we propose to adopt a full volume framework, which feeds the full volume brain image into the segmentation network and directly outputs the segmentation result for the whole brain volume. The framework makes use of complete information in each volume and can be implemented easily. An effective instance in this framework is given subsequently. We adopt the 3D high-resolution network (HRNet) for learning spatially fine-grained representations and the mixed precision training scheme for memory-efficient training. Extensive experiment results on a publicly available 3D MRI brain dataset show that our proposed model advances the state-of-the-art methods in terms of segmentation performance.

Content-aware convolutional neural networks.

Convolutional Neural Networks (CNNs) have achieved great success due to the powerful feature learning ability of convolution layers. Specifically, the standard convolution traverses the input images/features using a sliding window scheme to extract features. However, not all the windows contribute equally to the prediction results of CNNs. In practice, the convolutional operation on some of the windows (e.g., smooth windows that contain very similar pixels) can be very redundant and may introduce noises into the computation. Such redundancy may not only deteriorate the performance but also incur the unnecessary computational cost. Thus, it is important to reduce the computational redundancy of convolution to improve the performance. To this end, we propose a Content-aware Convolution (CAC) that automatically detects the smooth windows and applies a 1 ×1 convolutional kernel to replace the original large kernel. In this sense, we are able to effectively avoid the redundant computation on similar pixels. By replacing the standard convolution in CNNs with our CAC, the resultant models yield significantly better performance and lower computational cost than the baseline models with the standard convolution. More critically, we are able to dynamically allocate suitable computation resources according to the data smoothness of different images, making it possible for content-aware computation. Extensive experiments on various computer vision tasks demonstrate the superiority of our method over existing methods.

Novel Chiral Thiourea Derived from Hydroquinine and l-Phenylglycinol: An Effective Catalyst for Enantio- and Diastereoselective Aza-Henry Reaction.

A series of chiral thiourea bearing multiple H-bond donors derived from hydroquinine has been reported. The aza-Henry reaction of isatin-derived ketimines and long-chain nitroalkanes catalyzed by these chiral thioureas can achieve high enantioselectivity (78-99% ee) and excellent diastereoselectivity (up to 99:1). This work is the first report on long-chain nitroalkanes as substrates with excellent diastereoselectivity in metal-free catalytic systems.

Group Reidentification with Multigrained Matching and Integration.

The task of reidentifying groups of people under different camera views is an important yet less-studied problem. Group reidentification (Re-ID) is a very challenging task since it is not only adversely affected by common issues in traditional single-object Re-ID problems, such as viewpoint and human pose variations, but also suffers from changes in group layout and group membership. In this paper, we propose a novel concept of group granularity by characterizing a group image by multigrained objects: individual people and subgroups of two and three people within a group. To achieve robust group Re-ID, we first introduce multigrained representations which can be extracted via the development of two separate schemes, that is, one with handcrafted descriptors and another with deep neural networks. The proposed representation seeks to characterize both appearance and spatial relations of multigrained objects, and is further equipped with importance weights which capture variations in intragroup dynamics. Optimal group-wise matching is facilitated by a multiorder matching process which, in turn, dynamically updates the importance weights in iterative fashion. We evaluated three multicamera group datasets containing complex scenarios and large dynamics, with experimental results demonstrating the effectiveness of our approach.

Deep High-Resolution Representation Learning for Visual Recognition.

High-resolution representations are essential for position-sensitive vision problems, such as human pose estimation, semantic segmentation, and object detection. Existing state-of-the-art frameworks first encode the input image as a low-resolution representation through a subnetwork that is formed by connecting high-to-low resolution convolutions in series (e.g., ResNet, VGGNet), and then recover the high-resolution representation from the encoded low-resolution representation. Instead, our proposed network, named as High-Resolution Network (HRNet), maintains high-resolution representations through the whole process. There are two key characteristics: (i) Connect the high-to-low resolution convolution streams in parallel and (ii) repeatedly exchange the information across resolutions. The benefit is that the resulting representation is semantically richer and spatially more precise. We show the superiority of the proposed HRNet in a wide range of applications, including human pose estimation, semantic segmentation, and object detection, suggesting that the HRNet is a stronger backbone for computer vision problems. All the codes are available at https://github.com/HRNet.