相比于高效视频编码(high efficiency video coding,HEVC)标准,新一代编码标准多功能视频编码(versatile video coding,VVC)引入了很多新的技术,其中包括四叉树(quadtree,QT)和多类型树(multi-type tree,MTT)划分,MTT划分由HEVC中的QT...相比于高效视频编码(high efficiency video coding,HEVC)标准,新一代编码标准多功能视频编码(versatile video coding,VVC)引入了很多新的技术,其中包括四叉树(quadtree,QT)和多类型树(multi-type tree,MTT)划分,MTT划分由HEVC中的QT划分延伸而来。新划分方法提高了压缩效率,但导致编码时间急剧增加。为了降低编码复杂度,提出了一种结合深度学习方法和MTT方向早期判决的快速帧内编码算法。首先使用轻量级的卷积神经网络(convolutional neural network,CNN)对QT和部分MTT进行预测划分,其余MTT则采用提前预测MTT划分方向的方法作进一步的优化。实验结果表明,所提方法能够大幅降低编码复杂度,相比于原始编码器的编码时间减少了74.3%,且只有3.3%的码率损失,性能优于对比的方法。展开更多
为了提升高效视频编码(High Efficiency Video Coding,HEVC)帧内编码的实时性能,本文提出的方法利用了引入偶数边长与步长的卷积核以及自注意力机制的轻量级卷积网络来预测编码树单元(Coding Tree Unit,CTU)的帧内划分结构,从而减少了...为了提升高效视频编码(High Efficiency Video Coding,HEVC)帧内编码的实时性能,本文提出的方法利用了引入偶数边长与步长的卷积核以及自注意力机制的轻量级卷积网络来预测编码树单元(Coding Tree Unit,CTU)的帧内划分结构,从而减少了编码器对CTU进行四叉树递归遍历划分的编码时间。原始编码策略中粗模式决策通过基于残差经哈德曼变换的预测残差绝对值总和(Sum of Absolute Transformed Difference,SATD)的损失值来估计率失真优化过程中的率失真损失值来进行加速,但仍会耗费一定的编码时间。提出一种方法通过采样搜索的方式减少粗模式决策过程中计算的模式数,从35种模式降低到了18种模式,降低了粗模式决策过程中计算估计损失值的时间。由粗模式决策过程得到的较优的多个候选帧内模式来进行率失真优化,为了缩减粗模式决策需要计算的候选模式数,在候选模式列表中根据前后帧内预测角度模式的估计损失值的差距来筛选掉部分可能性较低的候选模式实现早停止决策,从而减少需要进行率失真优化的候选模式数量,进而减少率失真优化过程的计算时间。本文提出的算法在测试序列上平均实现78.15%的编码时间缩减,BD-PSNR为-0.168 d B,BD-RATE为3.49%。展开更多
在R-λ帧内码控中,提出基于卷积神经网络(Convolutional Neural Networks,CNN)的最佳比特分配和最优拉格朗日因子λ选择。首先,探索编码树单元(Coding Tree Unit,CTU)的码率与失真(Rate-Distortion,R-D)及码率与拉格朗日因子λ(Rate-λ...在R-λ帧内码控中,提出基于卷积神经网络(Convolutional Neural Networks,CNN)的最佳比特分配和最优拉格朗日因子λ选择。首先,探索编码树单元(Coding Tree Unit,CTU)的码率与失真(Rate-Distortion,R-D)及码率与拉格朗日因子λ(Rate-λ,R-λ)的关系特性,设计具有四输出的CNN预测R-D和R-λ曲线的关键参数;然后,建立帧级λ和目标码率的优化方程,反演得到最佳CTU码率分配;最后,根据CTU码率分配和先知的R-λ曲线,得到最优CTU级λ。实验表明,算法在保持4.76%控制精度下,比VTM13.0默认码控算法提高0.31 dB的编码质量。展开更多
The mortality rate of acute severe intraventricular hematoma is extremely high, and the rate of disability in survivors is high. Intraventricular hematoma has always been a difficult problem for clinical treatment. Al...The mortality rate of acute severe intraventricular hematoma is extremely high, and the rate of disability in survivors is high. Intraventricular hematoma has always been a difficult problem for clinical treatment. Although minimally invasive endoscopic hematoma evacuation is widely used to treat this disease, the technique still has room for improvement. Equipment for the intra-neuroendoscopic technique（INET） consists of two of our patented inventions： a transparent sheath（Patent No. ZL 200820046232.0） and a hematoma aspirator（Patent No. ZL 201520248717.8）. This study explored the safety and efficacy of INET by comparing it with extraventricular drainage in combination with urokinase thrombolytic therapy. This trial recruited 65 patients with severe intraventricular hemorrhage, including 35（19 men and 16 women, aged 53.2 ± 8.7 years） in the INET group and 30（17 men and 13 women, aged 51.5 ± 7.9 years） in the control group（extraventricular drainage plus urokinase thrombolytic therapy）. Our results showed that compared with the control group, the INET group exhibited lower intraventricular hemorrhage volumes, shorter intensive care-unit monitoring and ventricular drainage-tube placement times, and fewer incidences of intracranial infection, secondary bleeding, and mortality. Thus, the prognosis of survivors had improved remarkably. These findings indicate that INET is a safe and efficient new method for treating severe intraventricular hematoma. This trial was registered with Clinical Trials.gov（NCT02515903）.展开更多
AIM To highlight the salient magnetic resonance imaging(MRI) features of the intraneural ganglion cyst(INGC) of various peripheral nerves for their precise diagnosis and to differentiate them from other intra and extr...AIM To highlight the salient magnetic resonance imaging(MRI) features of the intraneural ganglion cyst(INGC) of various peripheral nerves for their precise diagnosis and to differentiate them from other intra and extraneural cystic lesions.METHODS A retrospective analysis of the magnetic resonance(MR) images of a cohort of 245 patients presenting with nerve palsy involving different peripheral nerves was done.MR images were analyzed for the presence of a nerve lesion,and if found,it was further characterized as solid or cystic.The serial axial,coronal and sagittal MR images of the lesions diagnosed as INGC were studied for their pattern and the anatomical extent along the course of the affected nerve and its branches.Its relation to identifiable anatomical landmarks,intraarticular communication and presence of denervation changes in the muscles supplied by involved nerve was also studied.RESULTS A total of 45 cystic lesions in the intra or extraneurallocations of the nerves were identified from the 245 MR scans done for patients presenting with nerve palsy.Out of these 45 cystic lesions,13 were diagnosed to have INGC of a peripheral nerve on MRI.The other cystic lesions included extraneural ganglion cyst,paralabral cyst impinging upon the suprascapular nerve,cystic schwannoma and nerve abscesses related to Hansen's disease involving various peripheral nerves.Thirteen lesions of INGC were identified in 12 patients.Seven of these affected the common peroneal nerve with one patient having a bilateral involvement.Two lesions each were noted in the tibial and suprascapular nerves,and one each in the obturator and proximal sciatic nerve.An intra-articular connection along the articular branch was demonstrated in 12 out of 13 lesions.Varying stages of denervation atrophy of the supplied muscles of the affected nerves were seen in 7 cases.Out of these 13 lesions in 12 patients,6 underwent surgery.CONCLUSION INGC is an important cause of reversible mono-neuropathy if diagnosed early and surgically treated.Its classic MRI pattern differentiates it from other lesions of the peripheral nerve and aid in its therapeutic planning.In each case,the joint connection has to be identified preoperatively,and the same should be excised during surgery to prevent further cyst recurrence.展开更多
近年来,卷积神经网络在实验室控制环境下的人脸表情识别任务中取得了很大进步,但是在自然场景中人脸表情识别方面仍然存在一些挑战.针对自然场景中人脸表情数据分布不平衡,以及由姿势、光照和性别等因素引起的类内差异大的问题,提出类...近年来,卷积神经网络在实验室控制环境下的人脸表情识别任务中取得了很大进步,但是在自然场景中人脸表情识别方面仍然存在一些挑战.针对自然场景中人脸表情数据分布不平衡,以及由姿势、光照和性别等因素引起的类内差异大的问题,提出类别均衡与局部中值(class-balanced and local median,CALM)损失函数.CALM损失函数包含类别均衡Softmax损失函数和局部中值损失函数2个部分.其中,类别均衡Softmax损失函数将数据量较少且容易错分的害怕和厌恶2种表情标记为难样本,将其余5种表情标记为易样本;在网络训练过程中对难样本自适应地增大权重,以提高难样本的识别准确率,进而提高表情识别的平均准确率.此外,在每个类别中会有一些离类别内大多数样本较远的样本,它们的存在会导致用均值方法计算出的类别中心偏离类内大多数样本.在局部中值损失函数中,采用与每个样本属于同类别的若干近邻的中值作为类别中心,在一定程度上减弱离群样本对类别中心选择的影响.在RAF(real-world affective faces)数据集上进行实验,与局部子类方法相比,该方法的平均识别准确率提升了1.32%,证明了该方法的有效性.展开更多
文摘相比于高效视频编码(high efficiency video coding,HEVC)标准,新一代编码标准多功能视频编码(versatile video coding,VVC)引入了很多新的技术,其中包括四叉树(quadtree,QT)和多类型树(multi-type tree,MTT)划分,MTT划分由HEVC中的QT划分延伸而来。新划分方法提高了压缩效率,但导致编码时间急剧增加。为了降低编码复杂度,提出了一种结合深度学习方法和MTT方向早期判决的快速帧内编码算法。首先使用轻量级的卷积神经网络(convolutional neural network,CNN)对QT和部分MTT进行预测划分,其余MTT则采用提前预测MTT划分方向的方法作进一步的优化。实验结果表明,所提方法能够大幅降低编码复杂度,相比于原始编码器的编码时间减少了74.3%,且只有3.3%的码率损失,性能优于对比的方法。
文摘为了提升高效视频编码(High Efficiency Video Coding,HEVC)帧内编码的实时性能,本文提出的方法利用了引入偶数边长与步长的卷积核以及自注意力机制的轻量级卷积网络来预测编码树单元(Coding Tree Unit,CTU)的帧内划分结构,从而减少了编码器对CTU进行四叉树递归遍历划分的编码时间。原始编码策略中粗模式决策通过基于残差经哈德曼变换的预测残差绝对值总和(Sum of Absolute Transformed Difference,SATD)的损失值来估计率失真优化过程中的率失真损失值来进行加速,但仍会耗费一定的编码时间。提出一种方法通过采样搜索的方式减少粗模式决策过程中计算的模式数,从35种模式降低到了18种模式,降低了粗模式决策过程中计算估计损失值的时间。由粗模式决策过程得到的较优的多个候选帧内模式来进行率失真优化,为了缩减粗模式决策需要计算的候选模式数,在候选模式列表中根据前后帧内预测角度模式的估计损失值的差距来筛选掉部分可能性较低的候选模式实现早停止决策,从而减少需要进行率失真优化的候选模式数量,进而减少率失真优化过程的计算时间。本文提出的算法在测试序列上平均实现78.15%的编码时间缩减,BD-PSNR为-0.168 d B,BD-RATE为3.49%。
基金funded by a grant from the Clinical Research Project of Shenzhen Health and Family Planning Commission in China,No.SZLY2018007a grant from the Science and Technology Planning Project of Shenzhen City of China,No.JCYJ 20150403101028210+2 种基金a grant from the Science and Technology Development and Cultivation Project of Southern Medical University of China,No.KJ 20161115the Guangdong Provincial Medical Research Fund in China,No.A2016545the Shenzhen Health Planning Commission Research Fund in China,No.201601013,201506009
文摘The mortality rate of acute severe intraventricular hematoma is extremely high, and the rate of disability in survivors is high. Intraventricular hematoma has always been a difficult problem for clinical treatment. Although minimally invasive endoscopic hematoma evacuation is widely used to treat this disease, the technique still has room for improvement. Equipment for the intra-neuroendoscopic technique（INET） consists of two of our patented inventions： a transparent sheath（Patent No. ZL 200820046232.0） and a hematoma aspirator（Patent No. ZL 201520248717.8）. This study explored the safety and efficacy of INET by comparing it with extraventricular drainage in combination with urokinase thrombolytic therapy. This trial recruited 65 patients with severe intraventricular hemorrhage, including 35（19 men and 16 women, aged 53.2 ± 8.7 years） in the INET group and 30（17 men and 13 women, aged 51.5 ± 7.9 years） in the control group（extraventricular drainage plus urokinase thrombolytic therapy）. Our results showed that compared with the control group, the INET group exhibited lower intraventricular hemorrhage volumes, shorter intensive care-unit monitoring and ventricular drainage-tube placement times, and fewer incidences of intracranial infection, secondary bleeding, and mortality. Thus, the prognosis of survivors had improved remarkably. These findings indicate that INET is a safe and efficient new method for treating severe intraventricular hematoma. This trial was registered with Clinical Trials.gov（NCT02515903）.
文摘AIM To highlight the salient magnetic resonance imaging(MRI) features of the intraneural ganglion cyst(INGC) of various peripheral nerves for their precise diagnosis and to differentiate them from other intra and extraneural cystic lesions.METHODS A retrospective analysis of the magnetic resonance(MR) images of a cohort of 245 patients presenting with nerve palsy involving different peripheral nerves was done.MR images were analyzed for the presence of a nerve lesion,and if found,it was further characterized as solid or cystic.The serial axial,coronal and sagittal MR images of the lesions diagnosed as INGC were studied for their pattern and the anatomical extent along the course of the affected nerve and its branches.Its relation to identifiable anatomical landmarks,intraarticular communication and presence of denervation changes in the muscles supplied by involved nerve was also studied.RESULTS A total of 45 cystic lesions in the intra or extraneurallocations of the nerves were identified from the 245 MR scans done for patients presenting with nerve palsy.Out of these 45 cystic lesions,13 were diagnosed to have INGC of a peripheral nerve on MRI.The other cystic lesions included extraneural ganglion cyst,paralabral cyst impinging upon the suprascapular nerve,cystic schwannoma and nerve abscesses related to Hansen's disease involving various peripheral nerves.Thirteen lesions of INGC were identified in 12 patients.Seven of these affected the common peroneal nerve with one patient having a bilateral involvement.Two lesions each were noted in the tibial and suprascapular nerves,and one each in the obturator and proximal sciatic nerve.An intra-articular connection along the articular branch was demonstrated in 12 out of 13 lesions.Varying stages of denervation atrophy of the supplied muscles of the affected nerves were seen in 7 cases.Out of these 13 lesions in 12 patients,6 underwent surgery.CONCLUSION INGC is an important cause of reversible mono-neuropathy if diagnosed early and surgically treated.Its classic MRI pattern differentiates it from other lesions of the peripheral nerve and aid in its therapeutic planning.In each case,the joint connection has to be identified preoperatively,and the same should be excised during surgery to prevent further cyst recurrence.
文摘近年来,卷积神经网络在实验室控制环境下的人脸表情识别任务中取得了很大进步,但是在自然场景中人脸表情识别方面仍然存在一些挑战.针对自然场景中人脸表情数据分布不平衡,以及由姿势、光照和性别等因素引起的类内差异大的问题,提出类别均衡与局部中值(class-balanced and local median,CALM)损失函数.CALM损失函数包含类别均衡Softmax损失函数和局部中值损失函数2个部分.其中,类别均衡Softmax损失函数将数据量较少且容易错分的害怕和厌恶2种表情标记为难样本,将其余5种表情标记为易样本;在网络训练过程中对难样本自适应地增大权重,以提高难样本的识别准确率,进而提高表情识别的平均准确率.此外,在每个类别中会有一些离类别内大多数样本较远的样本,它们的存在会导致用均值方法计算出的类别中心偏离类内大多数样本.在局部中值损失函数中,采用与每个样本属于同类别的若干近邻的中值作为类别中心,在一定程度上减弱离群样本对类别中心选择的影响.在RAF(real-world affective faces)数据集上进行实验,与局部子类方法相比,该方法的平均识别准确率提升了1.32%,证明了该方法的有效性.
