arXiv subjects

Peng Cao

Publications and source records attributed to Peng Cao.

51 records · Page 3Linked to original sources

arXiv2023

Edge Enhanced Image Style Transfer via Transformers

In recent years, arbitrary image style transfer has attracted more and more attention. Given a pair of content and style images, a stylized one is hoped that retains the content from the former while catching style patterns from the latter. However, it is difficult to simultaneously keep well the trade-off between the content details and the style features. To stylize the image with sufficient style patterns, the content details may be damaged and sometimes the objects of images can not be distinguished clearly. For this reason, we present a new transformer-based method named STT for image style transfer and an edge loss which can enhance the content details apparently to avoid generating blurred results for excessive rendering on style features. Qualitative and quantitative experiments demonstrate that STT achieves comparable performance to state-of-the-art image style transfer methods while alleviating the content leak problem.

cs.CV↗

arXiv2022

Contactless Oxygen Monitoring with Gated Transformer

With the increasing popularity of telehealth, it becomes critical to ensure that basic physiological signals can be monitored accurately at home, with minimal patient overhead. In this paper, we propose a contactless approach for monitoring patients' blood oxygen at home, simply by analyzing the radio signals in the room, without any wearable devices. We extract the patients' respiration from the radio signals that bounce off their bodies and devise a novel neural network that infers a patient's oxygen estimates from their breathing signal. Our model, called \emph{Gated BERT-UNet}, is designed to adapt to the patient's medical indices (e.g., gender, sleep stages). It has multiple predictive heads and selects the most suitable head via a gate controlled by the person's physiological indices. Extensive empirical results show that our model achieves high accuracy on both medical and radio datasets.

cs.LG↗

arXiv2022

Targeted Supervised Contrastive Learning for Long-Tailed Recognition

Real-world data often exhibits long tail distributions with heavy class imbalance, where the majority classes can dominate the training process and alter the decision boundaries of the minority classes. Recently, researchers have investigated the potential of supervised contrastive learning for long-tailed recognition, and demonstrated that it provides a strong performance gain. In this paper, we show that while supervised contrastive learning can help improve performance, past baselines suffer from poor uniformity brought in by imbalanced data distribution. This poor uniformity manifests in samples from the minority class having poor separability in the feature space. To address this problem, we propose targeted supervised contrastive learning (TSC), which improves the uniformity of the feature distribution on the hypersphere. TSC first generates a set of targets uniformly distributed on a hypersphere. It then makes the features of different classes converge to these distinct and uniformly distributed targets during training. This forces all classes, including minority classes, to maintain a uniform distribution in the feature space, improves class boundaries, and provides better generalization even in the presence of long-tail data. Experiments on multiple datasets show that TSC achieves state-of-the-art performance on long-tailed recognition tasks.

cs.CV↗

arXiv2022

A Timing Yield Model for SRAM Cells in Sub/Near-threshold Voltages Based on A Compact Drain Current Model

Sub/Near-threshold static random-access memory (SRAM) design is crucial for addressing the memory bottleneck in energy-constrained applications. However, the high integration density and reliability under process variations demand an accurate estimation of extremely small failure probabilities. To capture such a rare event in memory circuits, the time and storage overhead of conventional Monte Carlo (MC) simulations cannot be tolerated. On the other hand, classic analytical methods predicting failure probabilities from a physical expression become inaccurate in the sub/near-threshold region due to the assumed distribution or the oversimplified drain current model for nanoscale devices. This work first proposes a simple but efficient drain current model to describe the drain-induced barrier lowering effect at low voltages. Based on that, the probability density functions of the interest metrics in SRAM are derived. Two analytical models are then put forward to evaluate SRAM dynamic stabilities including access and write-time failures. The proposed models can be extended easily to different types of SRAM with different read/write assist circuits. The models are validated against MC simulations across different operating voltages and temperatures. The average relative errors at 0.5V VDD are only 8.8% and 10.4% for the access-time and write failure models respectively. The size of required data samples is 43.6X smaller than that of the state-of-the-art method.

cs.AR↗

arXiv2022

MRI Reconstruction Using Deep Bayesian Estimation

Purpose: To develop a deep learning-based Bayesian inference for MRI reconstruction. Methods: We modeled the MRI reconstruction problem with Bayes's theorem, following the recently proposed PixelCNN++ method. The image reconstruction from incomplete k-space measurement was obtained by maximizing the posterior possibility. A generative network was utilized as the image prior, which was computationally tractable, and the k-space data fidelity was enforced by using an equality constraint. The stochastic backpropagation was utilized to calculate the descent gradient in the process of maximum a posterior, and a projected subgradient method was used to impose the equality constraint. In contrast to the other deep learning reconstruction methods, the proposed one used the likelihood of prior as the training loss and the objective function in reconstruction to improve the image quality. Results: The proposed method showed an improved performance in preserving image details and reducing aliasing artifacts, compared with GRAPPA, $\ell_1$-ESPRiT, and MODL, a state-of-the-art deep learning reconstruction method. The proposed method generally achieved more than 5 dB peak signal-to-noise ratio improvement for compressed sensing and parallel imaging reconstructions compared with the other methods. Conclusion: The Bayesian inference significantly improved the reconstruction performance, compared with the conventional $\ell_1$-sparsity prior in compressed sensing reconstruction tasks. More importantly, the proposed reconstruction framework can be generalized for most MRI reconstruction scenarios.

cs.CV↗

arXiv2022

UCTransNet: Rethinking the Skip Connections in U-Net from a Channel-wise Perspective with Transformer

Most recent semantic segmentation methods adopt a U-Net framework with an encoder-decoder architecture. It is still challenging for U-Net with a simple skip connection scheme to model the global multi-scale context: 1) Not each skip connection setting is effective due to the issue of incompatible feature sets of encoder and decoder stage, even some skip connection negatively influence the segmentation performance; 2) The original U-Net is worse than the one without any skip connection on some datasets. Based on our findings, we propose a new segmentation framework, named UCTransNet (with a proposed CTrans module in U-Net), from the channel perspective with attention mechanism. Specifically, the CTrans module is an alternate of the U-Net skip connections, which consists of a sub-module to conduct the multi-scale Channel Cross fusion with Transformer (named CCT) and a sub-module Channel-wise Cross-Attention (named CCA) to guide the fused multi-scale channel-wise information to effectively connect to the decoder features for eliminating the ambiguity. Hence, the proposed connection consisting of the CCT and CCA is able to replace the original skip connection to solve the semantic gaps for an accurate automatic medical image segmentation. The experimental results suggest that our UCTransNet produces more precise segmentation performance and achieves consistent improvements over the state-of-the-art for semantic segmentation across different datasets and conventional architectures involving transformer or U-shaped framework. Code: https://github.com/McGregorWwww/UCTransNet.

cs.CV↗

arXiv2021

A Spatio-temporal Transformer for 3D Human Motion Prediction

We propose a novel Transformer-based architecture for the task of generative modelling of 3D human motion. Previous work commonly relies on RNN-based models considering shorter forecast horizons reaching a stationary and often implausible state quickly. Recent studies show that implicit temporal representations in the frequency domain are also effective in making predictions for a predetermined horizon. Our focus lies on learning spatio-temporal representations autoregressively and hence generation of plausible future developments over both short and long term. The proposed model learns high dimensional embeddings for skeletal joints and how to compose a temporally coherent pose via a decoupled temporal and spatial self-attention mechanism. Our dual attention concept allows the model to access current and past information directly and to capture both the structural and the temporal dependencies explicitly. We show empirically that this effectively learns the underlying motion dynamics and reduces error accumulation over time observed in auto-regressive models. Our model is able to make accurate short-term predictions and generate plausible motion sequences over long horizons. We make our code publicly available at https://github.com/eth-ait/motion-transformer.

cs.CV↗

arXiv2021

Deep Learning based Spectral CT Imaging

Spectral computed tomography (CT) has attracted much attention in radiation dose reduction, metal artifacts removal, tissue quantification and material discrimination. The x-ray energy spectrum is divided into several bins, each energy-bin-specific projection has a low signal-noise-ratio (SNR) than the current-integrating counterpart, which makes image reconstruction a unique challenge. Traditional wisdom is to use prior knowledge based iterative methods. However, this kind of methods demands a great computational cost. Inspired by deep learning, here we first develop a deep learning based reconstruction method; i.e., U-net with L_p^p-norm, Total variation, Residual learning, and Anisotropic adaption (ULTRA). Specifically, we emphasize the Various Multi-scale Feature Fusion and Multichannel Filtering Enhancement with a denser connection encoding architecture for residual learning and feature fusion. To address the image deblurring problem associated with the $L_2^2$-loss, we propose a general $L_p^p$-loss, $p>0$ Furthermore, the images from different energy bins share similar structures of the same object, the regularization characterizing correlations of different energy bins is incorporated into the $L_p^p$-loss function, which helps unify the deep learning based methods with traditional compressed sensing based methods. Finally, the anisotropically weighted total variation is employed to characterize the sparsity in the spatial-spectral domain to regularize the proposed network. In particular, we validate our ULTRA networks on three large-scale spectral CT datasets, and obtain excellent results relative to the competing algorithms. In conclusion, our quantitative and qualitative results in numerical simulation and preclinical experiments demonstrate that our proposed approach is accurate, efficient and robust for high-quality spectral CT image reconstruction.

eess.IV↗

arXiv2020

TCGM: An Information-Theoretic Framework for Semi-Supervised Multi-Modality Learning

Fusing data from multiple modalities provides more information to train machine learning systems. However, it is prohibitively expensive and time-consuming to label each modality with a large amount of data, which leads to a crucial problem of semi-supervised multi-modal learning. Existing methods suffer from either ineffective fusion across modalities or lack of theoretical guarantees under proper assumptions. In this paper, we propose a novel information-theoretic approach, namely \textbf{T}otal \textbf{C}orrelation \textbf{G}ain \textbf{M}aximization (TCGM), for semi-supervised multi-modal learning, which is endowed with promising properties: (i) it can utilize effectively the information across different modalities of unlabeled data points to facilitate training classifiers of each modality (ii) it has theoretical guarantee to identify Bayesian classifiers, i.e., the ground truth posteriors of all modalities. Specifically, by maximizing TC-induced loss (namely TC gain) over classifiers of all modalities, these classifiers can cooperatively discover the equivalent class of ground-truth classifiers; and identify the unique ones by leveraging limited percentage of labeled data. We apply our method to various tasks and achieve state-of-the-art results, including news classification, emotion recognition and disease prediction.

cs.CV↗

arXiv2020

Bone Material Analogues for PET/MRI Phantoms

Purpose: To develop bone material analogues that can be used in construction of phantoms for simultaneous PET/MRI systems. Methods: Plaster was used as the basis for the bone material analogues tested in this study. It was mixed with varying concentrations of an iodinated CT contrast, a gadolinium-based MR contrast agent, and copper sulfate to modulate the attenuation properties and MRI properties (T1 and T2*). Attenuation was measured with CT and 68Ge transmission scans, and MRI properties were measured with quantitative ultrashort echo time pulse sequences. A proof-of-concept skull was created by plaster casting. Results: Undoped plaster has a 511 keV attenuation coefficient (~0.14 cm-1) similar to cortical bone (0.10-0.15 cm-1), but slightly longer T1 (~500 ms) and T2* (~1.2 ms) MR parameters compared to bone (T1 ~ 300 ms, T2* ~ 0.4 ms). Doping with the iodinated agent resulted in increased attenuation with minimal perturbation to the MR parameters. Doping with a gadolinium chelate greatly reduced T1 and T2*, resulting in extremely short T1 values when the target T2* values were reached, while the attenuation coefficient was unchanged. Doping with copper sulfate was more selective for T2* shortening and achieved comparable T1 and T2* values to bone (after 1 week of drying), while the attenuation coefficient was unchanged. Conclusions: Plaster doped with copper sulfate is a promising bone material analogue for a PET/MRI phantom, mimicking the MR properties (T1 and T2*) and 511 keV attenuation coefficient of human cortical bone.

physics.med-ph↗

arXiv2020

Development of Conditional Random Field Insert for UNet-based Zonal Prostate Segmentation on T2-Weighted MRI

Purpose: A conventional 2D UNet convolutional neural network (CNN) architecture may result in ill-defined boundaries in segmentation output. Several studies imposed stronger constraints on each level of UNet to improve the performance of 2D UNet, such as SegNet. In this study, we investigated 2D SegNet and a proposed conditional random field insert (CRFI) for zonal prostate segmentation from clinical T2-weighted MRI data. Methods: We introduced a new methodology that combines SegNet and CRFI to improve the accuracy and robustness of the segmentation. CRFI has feedback connections that encourage the data consistency at multiple levels of the feature pyramid. On the encoder side of the SegNet, the CRFI combines the input feature maps and convolution block output based on their spatial local similarity, like a trainable bilateral filter. For all networks, 725 2D images (i.e., 29 MRI cases) were used in training; while, 174 2D images (i.e., 6 cases) were used in testing. Results: The SegNet with CRFI achieved the relatively high Dice coefficients (0.76, 0.84, and 0.89) for the peripheral zone, central zone, and whole gland, respectively. Compared with UNet, the SegNet+CRFIs segmentation has generally higher Dice score and showed the robustness in determining the boundaries of anatomical structures compared with the SegNet or UNet segmentation. The SegNet with a CRFI at the end showed the CRFI can correct the segmentation errors from SegNet output, generating smooth and consistent segmentation for the prostate. Conclusion: UNet based deep neural networks demonstrated in this study can perform zonal prostate segmentation, achieving high Dice coefficients compared with those in the literature. The proposed CRFI method can reduce the fuzzy boundaries that affected the segmentation performance of baseline UNet and SegNet models.

q-bio.QM↗

arXiv2019

Simultaneous Segmentation and Relaxometry for MRI through Multitask Learning

Purpose: This study demonstrated an MR signal multitask learning method for 3D simultaneous segmentation and relaxometry of human brain tissues. Materials and Methods: A 3D inversion-prepared balanced steady-state free precession sequence was used for acquiring in vivo multi-contrast brain images. The deep neural network contained 3 residual blocks, and each block had 8 fully connected layers with sigmoid activation, layer norm, and 256 neurons in each layer. Online synthesized MR signal evolutions and labels were used to train the neural network batch-by-batch. Empirically defined ranges of T1 and T2 values for the normal gray matter, white matter and cerebrospinal fluid (CSF) were used as the prior knowledge. MRI brain experiments were performed on 3 healthy volunteers as well as animal (N=6) and prostate patient (N=1) experiments. Results: In animal validation experiment, the differences/errors (mean difference $\pm$ standard deviation of difference) between the T1 and T2 values estimated from the proposed method and the ground truth were 113 $\pm$ 486 and 154 $\pm$ 512 ms for T1, and 5 $\pm$ 33 and 7 $\pm$ 41 ms for T2, respectively. In healthy volunteer experiments (N=3), whole brain segmentation and relaxometry were finished within ~5 seconds. The estimated apparent T1 and T2 maps were in accordance with known brain anatomy, and not affected by coil sensitivity variation. Gray matter, white matter, and CSF were successfully segmented. The deep neural network can also generate synthetic T1 and T2 weighted images. Conclusion: The proposed multitask learning method can directly generate brain apparent T1 and T2 maps, as well as synthetic T1 and T2 weighted images, in conjunction with segmentation of gray matter, white matter and CSF.

physics.med-ph↗

arXiv2019

L_DMI: An Information-theoretic Noise-robust Loss Function

Accurately annotating large scale dataset is notoriously expensive both in time and in money. Although acquiring low-quality-annotated dataset can be much cheaper, it often badly damages the performance of trained models when using such dataset without particular treatment. Various methods have been proposed for learning with noisy labels. However, most methods only handle limited kinds of noise patterns, require auxiliary information or steps (e.g. , knowing or estimating the noise transition matrix), or lack theoretical justification. In this paper, we propose a novel information-theoretic loss function, $\mathcal{L}_{DMI}$, for training deep neural networks robust to label noise. The core of $\mathcal{L}_{DMI}$ is a generalized version of mutual information, termed Determinant based Mutual Information (DMI), which is not only information-monotone but also relatively invariant. \emph{To the best of our knowledge, $\mathcal{L}_{DMI}$ is the first loss function that is provably robust to instance-independent label noise, regardless of noise pattern, and it can be applied to any existing classification neural networks straightforwardly without any auxiliary information}. In addition to theoretical justification, we also empirically show that using $\mathcal{L}_{DMI}$ outperforms all other counterparts in the classification task on both image dataset and natural language dataset include Fashion-MNIST, CIFAR-10, Dogs vs. Cats, MR with a variety of synthesized noise patterns and noise amounts, as well as a real-world dataset Clothing1M. Codes are available at https://github.com/Newbeeer/L_DMI .

cs.LG↗

arXiv2019

Deep convolution neural network model for automatic risk assessment of patients with non-metastatic nasopharyngeal carcinoma

Nasopharyngeal Carcinoma (NPC) is endemic cancer in the south-east Asia. With the advent of intensity-modulated radiotherapy excellent locoregional control are being achieved. Consequently, this had led to pretreatment clinical staging classification to be less prognostic of outcomes such as recurrence after treatment. Alternative pretreatment strategies for prognosis of NPC after treatment are needed to provide better risk stratification for NPC. In this study we proposed a deep convolution neural network model based on contrast-enhanced T1 (T1C) and T2 weighted (T2) MRI scan to predict 3-year disease progression of NPC patient after primary treatment. We retrospective obtained 596 non-metastatic NPC patients from four independent centres in Hong Kong and China. Our model first performs a segmentation of the primary NPC tumour to localise the tumour, and then uses the segmentation mask as prior knowledge along with the T1C and T2 scan to classify 3-year disease progression. For segmentation, we adapted and modified a VNet to encode both T1C and T2 scan and also encoding to classify T and overall stage classification. Our modified network performed better than baseline VNet with T1C and network with no T and overall classification. The classification result for 3-year disease progression achieved an AUC of 0.828 in the validation set but did not generalised well for the test set which consist of 146 patients from a different centre to the training data (AUC = 0.69). Our preliminary results show that deep learning may offer prognostication of disease progression of NPC patients after treatment. One advantage of our model is that it does not require manual segmentation of the region of interest, hence reducing clinician's burden. Further development in generalising multicentre data set are needed before clinical application of deep learning models in assessment of NPC.

eess.IV↗

arXiv2019

Max-MIG: an Information Theoretic Approach for Joint Learning from Crowds

Eliciting labels from crowds is a potential way to obtain large labeled data. Despite a variety of methods developed for learning from crowds, a key challenge remains unsolved: \emph{learning from crowds without knowing the information structure among the crowds a priori, when some people of the crowds make highly correlated mistakes and some of them label effortlessly (e.g. randomly)}. We propose an information theoretic approach, Max-MIG, for joint learning from crowds, with a common assumption: the crowdsourced labels and the data are independent conditioning on the ground truth. Max-MIG simultaneously aggregates the crowdsourced labels and learns an accurate data classifier. Furthermore, we devise an accurate data-crowds forecaster that employs both the data and the crowdsourced labels to forecast the ground truth. To the best of our knowledge, this is the first algorithm that solves the aforementioned challenge of learning from crowds. In addition to the theoretical validation, we also empirically show that our algorithm achieves the new state-of-the-art results in most settings, including the real-world data, and is the first algorithm that is robust to various information structures. Codes are available at \hyperlink{https://github.com/Newbeeer/Max-MIG}{https://github.com/Newbeeer/Max-MIG}

cs.LG↗