Search

Computer animation researchers have been extensively investigating 3D facial-expression synthesis for decades. However, flexible, robust production of realistic 3D facial expressions is still technically challenging. A proposed modeling framework applies sparse coding to synthesize 3D expressive faces, using specified coefficients or expression examples. It also robustly recovers facial expressions from noisy and incomplete data. This approach can synthesize higher-quality expressions in less time than the state-of-the-art techniques. [ABSTRACT FROM PUBLISHER]

Heterogeneous graph embedding aims at learning low-dimensional representations from a graph featuring nodes and edges of diverse natures, and meanwhile preserving the underlying topology. Existing approaches along this line have largely relied on meta-paths, which are by nature hand-crafted and pre-defined transition rules, so as to explore the semantics of a graph. Despite the promising results, defining meta-paths requires domain knowledge, and thus when the test distribution deviates from the priors, such methods are prone to errors. In this paper, we propose a self-learning scheme for heterogeneous graph embedding, termed as self-guided walk (SILK), that bypasses meta-paths and learns adaptive attentions for node walking. SILK assumes no prior knowledge or annotation is provided, and conducts a customized random walk to encode the contexts of the heterogeneous graph of interest. Specifically, this is achieved via maintaining a dynamically-updated guidance matrix that records the node-conditioned transition potentials. Experimental results on four real-world datasets demonstrate that SILK significantly outperforms state-of-the-art methods. [ABSTRACT FROM AUTHOR]

The goal of image steganography is to hide a full-sized image, termed secret, into another, termed cover. Prior image steganography algorithms can conceal only one secret within one cover. In this paper, we propose an adaptive local image steganography (AdaSteg) system that allows for scale- and location-adaptive image steganography. By adaptively hiding the secret on a local scale, the proposed system makes the steganography more secured, and further enables multi-secret steganography within one single cover. Specifically, this is achieved via two stages, namely the adaptive patch selection stage and secret encryption stage. Given a pair of secret and cover, first, the optimal local patch for concealment is determined adaptively by exploiting deep reinforcement learning with the proposed steganography quality function and policy network. The secret image is then converted into a patch of encrypted noises, resembling the process of generating adversarial examples, which are further encoded to a local region of the cover to realize a more secured steganography. Furthermore, we propose a novel criterion for the assessment of local steganography, and also collect a challenging dataset that is specialized for the task of image steganography, thus contributing to a standardized benchmark for the area. Experimental results demonstrate that the proposed model yields results superior to the state of the art in both security and capacity. [ABSTRACT FROM AUTHOR]

In this paper, we investigate regularized online exponentially concave (abbr. exp-concave) optimization, in which each loss function consists of a time-varying exp-concave function and a fixed convex regularization. If the whole loss function is exp-concave, a classical method called online Newton step (ONS) enjoys an O (d log T) regret bound, where d is the dimensionality and T is the time horizon. However, in the regularized setting, the sum of an exp-concave function and a convex regularization is not necessarily an exp-concave function, which implies that ONS is not applicable. To address this problem, we propose the proximal online Newton step (ProxONS), and show that it can attain the same O (d log T) regret bound for any convex regularization. The main idea is to first perform an iteration of ONS with the exp-concave part in each loss function and then perform a proximal mapping with the regularization part. Furthermore, we demonstrate that by utilizing the standard online-to-batch conversion, our ProxONS can be extended to solve stochastic optimization with a regularized exp-concave objective, and enjoy an O (d log T / T) convergence rate with high probability. Experimental results on two real datasets verify the effectiveness of our ProxONS. [ABSTRACT FROM AUTHOR]

Vision transformers have achieved remarkable success in computer vision tasks by using multi-head self-attention modules to capture long-range dependencies within images. However, the high inference computation cost poses a new challenge. Several methods have been proposed to address this problem, mainly by slimming patches. In the inference stage, these methods classify patches into two classes, one to keep and the other to discard in multiple layers. This approach results in additional computation at every layer where patches are discarded, which hinders inference acceleration. In this study, we tackle the patch slimming problem from a different perspective by proposing a life regression module that determines the lifespan of each image patch in one go. During inference, the patch is discarded once the current layer index exceeds its life. Our proposed method avoids additional computation and parameters in multiple layers to enhance inference speed while maintaining competitive performance. Additionally, our approach 1 1 https://github.com/cjwcommuny/life-regression. requires fewer training epochs than other patch slimming methods. [ABSTRACT FROM AUTHOR]

Traffic trajectories of various vehicles, bicycles and pedestrians can help understand the traffic dynamics in a fine-grained manner like traffic flow, traffic congestion and ride-hailing demand. The comprehensive usage of traffic trajectory data has not been fully investigated due to the prevalent privacy concerns and commercial limitations. The traffic trajectory simulation task has emerged to generate high-fidelity trajectories in demand for downstream tasks to fill the gap between the scarce trajectory data and the widespread applications. Previous state-of-the-art methods build the spatiotemporal dependencies of trajectories with Graph Neural Networks (GNNs) under generative adversarial training, yielding better yet unstable trajectory quality. We observe that the unsatisfied synthetic trajectories are caused by the insufficient spatiotemporal modeling of road networks and trajectory semantics. In this paper, we propose a novel S patio TE mporal GA ted (STEGA) framework equipped with semantic-aware graph learning for traffic trajectory simulation to enable the explicit modeling of spatiotemporal dependencies throughout the learning pipeline. On the one hand, STEGA employs a graph encoder with the semantics of road networks for the spatial points of a trajectory, together with a time encoder for the time points. On the other hand, STEGA devises two spatiotemporal gates with the semantic graphs for the predictions of the future trajectory. Boosted by the semantic-aware graph learning, the proposed STEGA outperforms the state-of-the-art counterparts consistently at both macro- and micro-level metrics on two datasets. Elaborate ablation studies and downstream tasks of the synthetic trajectories further demonstrate the superiority of STEGA. Our code is available at https://github.com/Star607/STEGA. • Semantic-aware graph learning is utilized throughout the entire pipeline. • Two spatiotemporal gates are devised with semantic graphs. • The proposed STEGA integrates GNN and powerful generative Transformer. • The proposed model outperforms all state-of-the-art baselines. [ABSTRACT FROM AUTHOR]

Object detection is one of the core tasks in computer vision that serves as a crucial underpinning for numerous applications. In recent years, deep learning-based methods have achieved remarkable performance in object detection. However, the performance of small objects still remains unsatisfactory. Therefore, some specific architectures have been proposed to address this issue in certain areas, such as remote-sensing and UAV images. In this paper, we aim to design a pluggable and non-intrusive method, termed as PatchDetector, to improve the performance of small object detection, which can effectively avoid the time and resource overhead of retraining the entire network. To achieve that, we first analyze why the mainstream networks perform poorly on small objects and find out that the fundamental reason is that the features of small are superseded by the background, which leads to a significant semantic gap in multi-level layers. Then, significance analysis is conducted to find the essential features for improving the small object detection. Next, with the located significant features, we devise a pluggable patch network for extracting essential features for small objects, which is non-intrusive to the original network. Experiments on mainstream detectors, including YOLO series and Faster RCNN, show that the proposed PatchDetector achieves 0. 4 % ∼ 2. 0 % mAP on small objects while not compromising the performance of medium and large objects. • The first pluggable and non-intrusive patch for small object detection performance. • We analyze the small object poor performance and locate the essential features. • Experiments demonstrate that the PatchDetector achieves 0. 4 % ∼ 2. 0 % mAP on small object. [ABSTRACT FROM AUTHOR]

Knowledge distillation (KD) has become a well established paradigm for compressing deep neural networks. The typical way of conducting knowledge distillation is to train the student network under the supervision of the teacher network to harness the knowledge at one or multiple spots (i.e., layers) in the teacher network. The distillation spots, once specified, will not change for all the training samples, throughout the whole distillation process. In this work, we argue that distillation spots should be adaptive to training samples and distillation epochs. We thus propose a new distillation strategy, termed spot-adaptive KD (SAKD), to adaptively determine the distillation spots in the teacher network per sample, at every training iteration during the whole distillation period. As SAKD actually focuses on “where to distill” instead of “what to distill” that is widely investigated by most existing works, it can be seamlessly integrated into existing distillation methods to further improve their performance. Extensive experiments with 10 state-of-the-art distillers are conducted to demonstrate the effectiveness of SAKD for improving their distillation performance, under both homogeneous and heterogeneous distillation settings. Code is available at https://github.com/zju-vipa/spot-adaptive-pytorch. [ABSTRACT FROM AUTHOR]

The dual-stream architecture is frequently employed for learning diverse features from videos. This paper introduces a novel Mixed Resolution Network (MixRes) for processing inputs with hybrid spatiotemporal resolutions, namely high-spatial and low-temporal resolution input, as well as low-spatial and high-temporal resolution input. The utilization of mixed spatiotemporal resolutions not only facilitates the independent emphasis of the two streams on appearance and motion encoding but also diminishes the computational burden. Furthermore, by leveraging the characteristics of neural networks with multiple layers, the temporal stream in the proposed network is divided into different steps to capture short-term and long-term motion information. Finally, we design a Temporal Multiscale Motion Excitation (TMME) module, which enhances the motion-related channels of the video representation by employing multiscale temporal differences. We conduct extensive experiments on multiple action recognition benchmarks, including Something-Something V1 & V2 and Kinetics-400. The outcomes validate that the proposed method achieves superior action recognition performance with low computational cost as compared to the state-of-the-art methods. [ABSTRACT FROM AUTHOR]

The misinformation prevention (MP) problem, which aims to find anchor nodes (P-seeds) to start an anti-spread of positive information, has attracted abundant research attention due to its practical application in reducing the risk of social unrest. Existing MP researches ideally assume that the M-seeds (nodes start misinformation spread) are known in advance. However, the misinformation spread usually occurs abruptly so that M-seeds are not easy to be identified as prior knowledge. In this paper, we study the MP problem with uncertain M-seeds. We assume that only a set of suspicious nodes (S-nodes) is known. Each S-node has a transition probability (T-prob) for becoming an M-seed. With different T-prob availability, we propose and study two new research problems: (1) the probabilistic MP (PMP) problem given exactly all the T-probs, and (2) the robust MP (RMP) problem given a confidence interval for each T-prob. For the PMP problem, we propose the prefixed Hybrid sampling based Misinformation Prevention (pHMP) algorithm that return a (1 − 1 / e − ɛ) -approximate solution. For the RMP problem, we optimize the worst-case ratio between the current solution and the optimal one. Two bi-criteria algorithms are proposed that achieve both δ absolute error and (1 − 1 / e − ɛ) 2 (1 − η) and (1 − 1 / e − ɛ) (1 − η) relative error respectively. We conduct extensive experiments on several real world datasets. The results show the proposed algorithms are superior over the baseline algorithm on all datasets, and robust against the changing of different application conditions. [ABSTRACT FROM AUTHOR]

We aimed to develop an explainable and reliable method to diagnose cysts and tumors of the jaw with massive panoramic radiographs of healthy peoples based on deep learning, since collecting and labeling massive lesion samples are time-consuming, and existing deep learning-based methods lack explainability. Based on the collected 872 lesion samples and 10,000 healthy samples, a two-branch network was proposed for classifying the cysts and tumors of the jaw. The two-branch network is firstly pretrained on massive panoramic radiographs of healthy peoples, then is trained for classifying the sample categories and segmenting the lesion area. Totally, 200 healthy samples and 87 lesion samples were included in the testing stage. The average accuracy, precision, sensitivity, specificity, and F1 score of classification are 88.72%, 65.81%, 66.56%, 92.66%, and 66.14%, respectively. The average accuracy, precision, sensitivity, specificity, and F1 score of classification will reach 90.66%, 85.23%, 84.27%, 93.50%, and 84.74%, if only classifying the lesion samples and healthy samples. The proposed method showed encouraging performance in the diagnosis of cysts and tumors of the jaw. The classified categories and segmented lesion areas serve as the diagnostic basis for further diagnosis, which provides a reliable tool for diagnosing jaw tumors and cysts. [ABSTRACT FROM AUTHOR]

• We introduce the disassembling task and an unsupervised approach towards solving it. • UDOR is devised to comply with the modularity of the learned latent representation. • We also introduce two metrics for evaluating the disassembling performance. In this paper, we study a new representation-learning task, which we termed as disassembling object representations. Given an image featuring multiple objects, the goal of disassembling is to acquire a latent representation, of which each part corresponds to one category of objects. Disassembling thus finds its application in a wide domain such as image editing and few- or zero-shot learning, as it enables category-specific modularity in the learned representations. To this end, we propose an unsupervised approach to achieving disassembling, named Unsupervised Disassembling Object Representation (UDOR). UDOR follows a double auto-encoder architecture, in which a fuzzy classification and an object-removing operation are imposed. The fuzzy classification constrains each part of the latent representation to encode features of up to one object category, while the object-removing, combined with a generative adversarial network, enforces the modularity of the representations and integrity of the reconstructed image. Furthermore, we devise two metrics to respectively measure the modularity of disassembled representations and the visual integrity of reconstructed images. Experimental results demonstrate that the proposed UDOR, despite unsupervised, achieves truly encouraging results on par with those of supervised methods. [ABSTRACT FROM AUTHOR]

The influence maximization problem asks to find a small number of early adopters of a product in a social network, such that the expected number of total adoptions is maximized over the network. The problem has been well-studied, but most of the studies focus on the case of a single product or purely competitive products. This paper proposes a new influence diffusion model for multiple complementary products, namely, the comparative linear threshold (Com-LT) model. Under the Com-LT model, we model the complementary relation by reducing the thresholds of nodes. With this model, we study two problems: SelfInfMax and CompInfMax. We prove that these two problems are both NP-Hard under the Com-LT model. For both the SelfInfMax and the CompInfMax problem, we theoretically analyze the monotonicity and submodularity, and accordingly leverage lower bound optimization to devise non-trivial effective approximation algorithms. We conduct experiments over 4 real-world datasets. The experimental results demonstrate the correctness and efficiency of the proposed algorithms. • We study the complementary version of the influence maximization problem. • We propose Com-LT model to describe the influence spread of non- competitive products. • We studied the SelfInfMax and the CompInfMax problem and show their hardness. • We propose approximation algorithms with high quality feasible solutions. • We conduct experiments on real networks and show the effectiveness of our algorithms. [ABSTRACT FROM AUTHOR]

Recently, many benchmark datasets have been found to contain noisy labels caused by unavoidable human mistakes. Many researchers propose new noise-aware loss functions to achieve robust classification performance. However, we discover that existing noise-aware loss functions cannot fully heal the damage caused by the noise. On the other hand, some methods filter out low confidence samples and train new models, whereas the filtered samples contain both noisy and hard samples that are critical for the robustness of models. Based on the above two discoveries, we devised the Noise-aware Network (NA-Net) for robust training with noisy labels. Each layer of NA-Net contains three groups of convolution kernels responsible for mix samples, clean samples, and noisy samples, termed as mix-kernels, clean-kernels, and noise-kernels, respectively. Mix-kernels are used for finding the clean samples with a newly devised noise-immune (NI) loss function; clean-kernels are targeted at learning better features without being misguided by noise; noise-kernels are trained by the remaining samples to rectify wrong labels for the next iteration. Meanwhile, for increasing the classification performance of mix-kernels, the extracted feature maps of clean-kernels without being poisoned are combined as the input of mix-kernels of the next layer. Also, the knowledge distillation strategy is adopted to distill the knowledge from clean-kernels to the noise-kernels. Extensive experiments demonstrate that the mutual promotion of three groups of kernels in NA-Net achieves state-of-the-art performance on both artificial noisy datasets and real-world datasets. • The first to clarify the influence extent of noisy and hard samples for training. • Noise-aware Network with separate groups of kernels is proposed for handling noise. • A new loss function is proposed to make the training robust to the noise. [ABSTRACT FROM AUTHOR]

Image-based nondestructive textile fiber identification is a challenging computer vision problem, that is practically useful in fashion, decoration, and design. Although deep learning now outperforms humans in many scenarios such as face and object recognition, image-based fiber identification is still an open problem for deep learning given imbalanced sample and small sample size samples. In this paper, we propose the Component Unmixing Network (CU-Net) for nondestructive textile fiber identification. CU-Net learns effective representations given imbalanced sample and small sample size samples to achieve high-performance textile fiber identification. CU-Net comprises a Deep Feature Extraction Module (DFE-Module) and a Component Unmixing Module (CU-Module). Initially, mixed deep features are extracted by DFE-Module from the input textile patches. Then, CU-Module is employed to extract unmixed representations of different fibers from the mixed deep features. In CU-Module, we introduce a self-interchange and a restraining loss to reduce the mixture between representations of different fibers. Furthermore, we extend CU-Net to the proportion analysis task with very good effect. Extensive experiments demonstrate that: (1) self-interchange and the restraining loss effectively unmix different fiber representations and improve fiber identification accuracy; and (2) CU-Net achieves more accurate fiber identification than the current state-of-the-art multi-label classification methods. [ABSTRACT FROM AUTHOR]

In this paper, we strive to propose a self-interpretable framework, termed PrimitiveTree , that incorporates deep visual primitives condensed from deep features with a conventional decision tree, bridging the gap between deep features extracted from deep neural networks (DNNs) and trees' transparent decision-making processes. Specifically, we utilize a codebook, which embeds the continuous deep features into a finite discrete space (deep visual primitives) to distill the most common semantic information. The decision tree adopts the spatial location information and the mapped primitives to present the decision-making process of the deep features in a tree hierarchy. Moreover, the trained interpretable PrimitiveTree can inversely explain the constituents of the deep features, highlighting the most critical and semantic-rich image patches attributing to the final predictions of the given DNN. Extensive experiments and visualization results validate the effectiveness and interpretability of our method. [Display omitted] • A codebook distills semantic-rich visual primitives from the continuous feature space. • The proposed PrimitiveTree can interpret deep features and reveal key image constituents. • Extensive experiments on popular datasets demonstrate the effectiveness and interpretability of our method. [ABSTRACT FROM AUTHOR]

Deep cooperative multi-agent reinforcement learning has demonstrated its remarkable success over a wide spectrum of complex control tasks. However, recent advances in multi-agent learning mainly focus on value decomposition while leaving entity interactions still intertwined, which easily leads to over-fitting on noisy interactions between entities. In this work, we introduce a novel interactiOn Pattern disenTangling (OPT) method, to disentangle the entity interactions into interaction prototypes, each of which represents an underlying interaction pattern within a subgroup of the entities. OPT facilitates filtering the noisy interactions between irrelevant entities and thus significantly improves generalizability as well as interpretability. Specifically, OPT introduces a sparse disagreement mechanism to encourage sparsity and diversity among discovered interaction prototypes. Then the model selectively restructures these prototypes into a compact interaction pattern by an aggregator with learnable weights. To alleviate the training instability issue caused by partial observability, we propose to maximize the mutual information between the aggregation weights and the history behaviors of each agent. Experiments on single-task, multi-task and zero-shot benchmarks demonstrate that the proposed method yields results superior to the state-of-the-art counterparts. Our code is available at https://github.com/liushunyu/OPT.

Researchers of temporal networks (e.g., social networks and transaction networks) have been interested in mining dynamic patterns of nodes from their diverse interactions. Inspired by recently powerful graph mining methods like skip-gram models and graph neural networks (GNNs), existing approaches focus on generating temporal node embeddings sequentially with nodes' sequential interactions. However, the sequential modeling of previous approaches cannot handles the transition structure between nodes' neighbors with limited memorization capacity. In detail, an effective method for the transition structures is required to both model nodes' personalized patterns adaptively and capture node dynamics accordingly. In this article, we propose a method, namely t ransition p ropagation g raph n eural n etworks (TIP-GNN), to tackle the challenges of encoding nodes' transition structures. The proposed TIP-GNN focuses on the bilevel graph structure in temporal networks: besides the explicit interaction graph, a node's sequential interactions can also be constructed as a transition graph. Based on the bilevel graph, TIP-GNN further encodes transition structures by multistep transition propagation and distills information from neighborhoods by a bilevel graph convolution. Experimental results over various temporal networks reveal the efficiency of our TIP-GNN, with at most 7.2% improvements of accuracy on temporal link prediction. Extensive ablation studies further verify the effectiveness and limitations of the transition propagation module. Our code is available at https://github.com/doujiang-zheng/TIP-GNN.

Background and Objective: Training a robust cancer diagnostic or prognostic artificial intelligent model using histology images requires a large number of representative cases with labels or annotations, which are difficult to obtain. The histology snapshots available in published papers or case reports can be used to enrich the training dataset. However, the magnifications of these invaluable snapshots are generally unknown, which limits their usage. Therefore, a robust magnification predictor is required for utilizing those diverse snapshot repositories consisting of different diseases. This paper presents a magnification prediction model named Hagnifinder for H&E-stained histological images., Methods: Hagnifinder is a regression model based on a modified convolutional neural network (CNN) that contains 3 modules: Feature Extraction Module, Regression Module, and Adaptive Scaling Module (ASM). In the training phase, the Feature Extraction Module first extracts the image features. Secondly, the ASM is proposed to address the learned feature values uneven distribution problem. Finally, the Regression Module estimates the mapping between the regularized extracted features and the magnifications. We construct a new dataset for training a robust model, named Hagni40, consisting of 94 643 H&E-stained histology image patches at 40 different magnifications of 13 types of cancer based on The Cancer Genome Atlas. To verify the performance of the Hagnifinder, we measure the accuracy of the predictions by setting the maximum allowable difference values (0.5, 1, and 5) between the predicted magnification and the actual magnification. We compare Hagnifinder with state-of-the-art methods on a public dataset BreakHis and the Hagni40., Results: The Hagnifinder provides consistent prediction accuracy, with a mean accuracy of 98.9%, across 40 different magnifications and 13 different cancer types when Resnet50 is used as the feature extractor. Compared with the state-of-the-art methods focusing on 4-5 levels of magnification classification, the Hagnifinder achieve s the best and most comparable performance in the BreakHis and Hagni40 datasets., Conclusions: The experimental results suggest that Hagnifinder can be a valuable tool for predicting the associated magnification of any given histology image., (© 2023 The Authors.)

Pseudo supervision is regarded as the core idea in semi-supervised learning for semantic segmentation, and there is always a tradeoff between utilizing only the high-quality pseudo labels and leveraging all the pseudo labels. Addressing that, we propose a novel learning approach, called Conservative-Progressive Collaborative Learning (CPCL), among which two predictive networks are trained in parallel, and the pseudo supervision is implemented based on both the agreement and disagreement of the two predictions. One network seeks common ground via intersection supervision and is supervised by the high-quality labels to ensure a more reliable supervision, while the other network reserves differences via union supervision and is supervised by all the pseudo labels to keep exploring with curiosity. Thus, the collaboration of conservative evolution and progressive exploration can be achieved. To reduce the influences of the suspicious pseudo labels, the loss is dynamic re-weighted according to the prediction confidence. Extensive experiments demonstrate that CPCL achieves state-of-the-art performance for semi-supervised semantic segmentation.

Knowledge amalgamation (KA) is a novel deep model reusing task aiming to transfer knowledge from several well-trained teachers to a multi-talented and compact student. Currently, most of these approaches are tailored for convolutional neural networks (CNNs). However, there is a tendency that Transformers, with a completely different architecture, are starting to challenge the domination of CNNs in many computer vision tasks. Nevertheless, directly applying the previous KA methods to Transformers leads to severe performance degradation. In this work, we explore a more effective KA scheme for Transformer-based object detection models. Specifically, considering the architecture characteristics of Transformers, we propose to dissolve the KA into two aspects: sequence-level amalgamation (SA) and task-level amalgamation (TA). In particular, a hint is generated within the sequence-level amalgamation by concatenating teacher sequences instead of redundantly aggregating them to a fixed-size one as previous KA approaches. Besides, the student learns heterogeneous detection tasks through soft targets with efficiency in the task-level amalgamation. Extensive experiments on PASCAL VOC and COCO have unfolded that the sequence-level amalgamation significantly boosts the performance of students, while the previous methods impair the students. Moreover, the Transformer-based students excel in learning amalgamated knowledge, as they have mastered heterogeneous detection tasks rapidly and achieved superior or at least comparable performance to those of the teachers in their specializations.

Chemical reaction prediction, involving forward synthesis and retrosynthesis prediction, is a fundamental problem in organic synthesis. A popular computational paradigm formulates synthesis prediction as a sequence-to-sequence translation problem, where the typical SMILES is adopted for molecule representations. However, the general-purpose SMILES neglects the characteristics of chemical reactions, where the molecular graph topology is largely unaltered from reactants to products, resulting in the suboptimal performance of SMILES if straightforwardly applied. In this article, we propose the root-aligned SMILES (R-SMILES), which specifies a tightly aligned one-to-one mapping between the product and the reactant SMILES for more efficient synthesis prediction. Due to the strict one-to-one mapping and reduced edit distance, the computational model is largely relieved from learning the complex syntax and dedicated to learning the chemical knowledge for reactions. We compare the proposed R-SMILES with various state-of-the-art baselines and show that it significantly outperforms them all, demonstrating the superiority of the proposed method., Competing Interests: There are no conflicts to declare., (This journal is © The Royal Society of Chemistry.)

Background: An accurate pathological diagnosis of hepatocellular carcinoma (HCC), one of the malignant tumors with the highest mortality rate, is time-consuming and heavily reliant on the experience of a pathologist. In this report, we proposed a deep learning model that required minimal noise reduction or manual annotation by an experienced pathologist for HCC diagnosis and classification., Methods: We collected a whole-slide image of hematoxylin and eosin-stained pathological slides from 592 HCC patients at the First Affiliated Hospital, College of Medicine, Zhejiang University between 2015 and 2020. We propose a noise-specific deep learning model. The model was trained initially with 137 cases cropped into multiple-scaled datasets. Patch screening and dynamic label smoothing strategies are adopted to handle the histopathological liver image with noise annotation from the perspective of input and output. The model was then tested in an independent cohort of 455 cases with comparable tumor types and differentiations., Results: Exhaustive experiments demonstrated that our two-step method achieved 87.81% pixel-level accuracy and 98.77% slide-level accuracy in the test dataset. Furthermore, the generalization performance of our model was also verified using The Cancer Genome Atlas dataset, which contains 157 HCC pathological slides, and achieved an accuracy of 87.90%., Conclusions: The noise-specific histopathological classification model of HCC based on deep learning is effective for the dataset with noisy annotation, and it significantly improved the pixel-level accuracy of the regular convolutional neural network (CNN) model. Moreover, the model also has an advantage in detecting well-differentiated HCC and microvascular invasion., Competing Interests: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The editor JC has declared a shared parent affiliation with the authors at the time of review., (Copyright © 2021 Feng, Yu, Liang, Li, Zhong, Hu, Zhang, Feng, Song, Zhang and Zhang.)

Human parsing and matting play important roles in various applications, such as dress collocation, clothing recommendation, and image editing. In this paper, we propose a lightweight hybrid model that unifies the fully-supervised hierarchical-granularity parsing task and the unsupervised matting one. Our model comprises two parts, the extensible hierarchical semantic segmentation block using CNN and the matting module composed of guided filters. Given a human image, the segmentation block stage-1 first obtains a primitive segmentation map to separate the human and the background. The primitive segmentation is then fed into stage-2 together with the original image to give a rough segmentation of human body. This procedure is repeated in the stage-3 to acquire a refined segmentation. The matting module takes as input the above estimated segmentation maps and produces the matting map, in a fully unsupervised manner. The obtained matting map is then in turn fed back to the CNN in the first block for refining the semantic segmentation results.

Both commercial and scientific applications often need to transform color images into gray-scale images, e.g., to reduce the publication cost in printing color images or to help color blind people see visual cues of color images. However, conventional color to gray algorithms are not ready for practical applications because they encounter the following problems: 1) Visual cues are not well defined so it is unclear how to preserve important cues in the transformed gray-scale images; 2) some algorithms have extremely high time cost for computation; and 3) some require human-computer interactions to have a reasonable transformation. To solve or at least reduce these problems, we propose a new algorithm based on a probabilistic graphical model with the assumption that the image is defined over a Markov random field. Thus, color to gray procedure can be regarded as a labeling process to preserve the newly well--defined visual cues of a color image in the transformed gray-scale image. Visual cues are measurements that can be extracted from a color image by a perceiver. They indicate the state of some properties of the image that the perceiver is interested in perceiving. Different people may perceive different cues from the same color image and three cues are defined in this paper, namely, color spatial consistency, image structure information, and color channel perception priority. We cast color to gray as a visual cue preservation procedure based on a probabilistic graphical model and optimize the model based on an integral minimization problem. We apply the new algorithm to both natural color images and artificial pictures, and demonstrate that the proposed approach outperforms representative conventional algorithms in terms of effectiveness and efficiency. In addition, it requires no human-computer interactions.