Jing Li

Hao Sun, Li Shen, Qihuang Zhong et al.

Sharpness aware minimization (SAM) optimizer has been extensively explored as it can generalize better for training deep neural networks via introducing extra perturbation steps to flatten the landscape of deep learning models. Integrating SAM with adaptive learning rate and momentum acceleration, dubbed AdaSAM, has already been explored empirically to train large-scale deep neural networks without theoretical guarantee due to the triple difficulties in analyzing the coupled perturbation step, adaptive learning rate and momentum step. In this paper, we try to analyze the convergence rate of AdaSAM in the stochastic non-convex setting. We theoretically show that AdaSAM admits a $\mathcal{O}(1/\sqrt{bT})$ convergence rate, which achieves linear speedup property with respect to mini-batch size $b$. Specifically, to decouple the stochastic gradient steps with the adaptive learning rate and perturbed gradient, we introduce the delayed second-order momentum term to decompose them to make them independent while taking an expectation during the analysis. Then we bound them by showing the adaptive learning rate has a limited range, which makes our analysis feasible. To the best of our knowledge, we are the first to provide the non-trivial convergence rate of SAM with an adaptive learning rate and momentum acceleration. At last, we conduct several experiments on several NLP tasks, which show that AdaSAM could achieve superior performance compared with SGD, AMSGrad, and SAM optimizers.

Jing Li, Zhengqi Zhang

We propose transformed Diffsuion-Wave fractional Physics-Informed Neural Networks (tDWfPINNs) for efficiently solving time-fractional diffusion-wave equations with fractional order $α\in(1,2)$. Conventional numerical methods for these equations often compromise the mesh-free advantage of Physics-Informed Neural Networks (PINNs) or impose high computational costs when computing fractional derivatives. The proposed method avoids first-order derivative calculations at quadrature points by introducing an integrand transformation technique, significantly reducing computational costs associated with fractional derivative evaluation while preserving accuracy. We conduct a comprehensive comparative analysis applying this integrand transformation in conjunction with both Monte Carlo integration and Gauss-Jacobi quadrature schemes across various time-fractional PDEs. Our results demonstrate that tDWfPINNs achieve superior computational efficiency without sacrificing accuracy. Furthermore, we incorporate the proposed approach into adaptive sampling approaches such as the residual-based adaptive distribution (RAD) for the time-fractional Burgers equation with order $α\in(1,2)$, which exhibits complex solution dynamics. The experiments show that the Gauss-Jacobi method typically outperforms the Monte Carlo approach; however, careful consideration is required when selecting the number of quadrature points. Overall, the proposed tDWfPINNs offer a significant advancement in the numerical solution of time-fractional diffusion-wave equations, providing an accurate and scalable mesh-free alternative for challenging fractional models.

Jing Li, Lu Bai, Bin Yang et al.

Infrared and visible image fusion aims to extract complementary features to synthesize a single fused image. Many methods employ convolutional neural networks (CNNs) to extract local features due to its translation invariance and locality. However, CNNs fail to consider the image's non-local self-similarity (NLss), though it can expand the receptive field by pooling operations, it still inevitably leads to information loss. In addition, the transformer structure extracts long-range dependence by considering the correlativity among all image patches, leading to information redundancy of such transformer-based methods. However, graph representation is more flexible than grid (CNN) or sequence (transformer structure) representation to address irregular objects, and graph can also construct the relationships among the spatially repeatable details or texture with far-space distance. Therefore, to address the above issues, it is significant to convert images into the graph space and thus adopt graph convolutional networks (GCNs) to extract NLss. This is because the graph can provide a fine structure to aggregate features and propagate information across the nearest vertices without introducing redundant information. Concretely, we implement a cascaded NLss extraction pattern to extract NLss of intra- and inter-modal by exploring interactions of different image pixels in intra- and inter-image positional distance. We commence by preforming GCNs on each intra-modal to aggregate features and propagate information to extract independent intra-modal NLss. Then, GCNs are performed on the concatenate intra-modal NLss features of infrared and visible images, which can explore the cross-domain NLss of inter-modal to reconstruct the fused image. Ablation studies and extensive experiments illustrates the effectiveness and superiority of the proposed method on three datasets.

Xianzhi Du, Yeqing Li, Yin Cui et al.

A recent work from Bello shows that training and scaling strategies may be more significant than model architectures for visual recognition. This short note studies effective training and scaling strategies for video recognition models. We propose a simple scaling strategy for 3D ResNets, in combination with improved training strategies and minor architectural changes. The resulting models, termed 3D ResNet-RS, attain competitive performance of 81.0 on Kinetics-400 and 83.8 on Kinetics-600 without pre-training. When pre-trained on a large Web Video Text dataset, our best model achieves 83.5 and 84.3 on Kinetics-400 and Kinetics-600. The proposed scaling rule is further evaluated in a self-supervised setup using contrastive learning, demonstrating improved performance. Code is available at: https://github.com/tensorflow/models/tree/master/official.

Abdullah Rashwan, Xianzhi Du, Xiaoqi Yin et al.

Scale-permuted networks have shown promising results on object bounding box detection and instance segmentation. Scale permutation and cross-scale fusion of features enable the network to capture multi-scale semantics while preserving spatial resolution. In this work, we evaluate this meta-architecture design on semantic segmentation - another vision task that benefits from high spatial resolution and multi-scale feature fusion at different network stages. By further leveraging dilated convolution operations, we propose SpineNet-Seg, a network discovered by NAS that is searched from the DeepLabv3 system. SpineNet-Seg is designed with a better scale-permuted network topology with customized dilation ratios per block on a semantic segmentation task. SpineNet-Seg models outperform the DeepLabv3/v3+ baselines at all model scales on multiple popular benchmarks in speed and accuracy. In particular, our SpineNet-S143+ model achieves the new state-of-the-art on the popular Cityscapes benchmark at 83.04% mIoU and attained strong performance on the PASCAL VOC2012 benchmark at 85.56% mIoU. SpineNet-Seg models also show promising results on a challenging Street View segmentation dataset. Code and checkpoints will be open-sourced.

Xianming Li, Jing Li

Social media data exhibits severe redundancy caused by its noisy nature. It leads to increased training time and model bias in its processing. To address this issue, we propose a novel Generative Deduplication framework for social media data selection by removing semantically duplicate data. While related work involves data selection in task-specific training, our model acts as an efficient pre-processing method to universally enhance social media NLP pipelines. Specifically, we train a generative model via self-supervised learning to predict a keyword to capture the semantics of noisy social media text for deduplication. Meanwhile, time-dimensional Gaussian noise is added to improve training complexity and avoid learning trivial features. Extensive experiments suggest that our model can better reduce training samples while improving performance than baselines. The results show our model's potential to broadly advance social media language understanding in effectiveness and efficiency.

Jing Li, Lu Bai, Bin Yang et al.

Infrared and visible image fusion (IVF) plays an important role in intelligent transportation system (ITS). The early works predominantly focus on boosting the visual appeal of the fused result, and only several recent approaches have tried to combine the high-level vision task with IVF. However, they prioritize the design of cascaded structure to seek unified suitable features and fit different tasks. Thus, they tend to typically bias toward to reconstructing raw pixels without considering the significance of semantic features. Therefore, we propose a novel prior semantic guided image fusion method based on the dual-modality strategy, improving the performance of IVF in ITS. Specifically, to explore the independent significant semantic of each modality, we first design two parallel semantic segmentation branches with a refined feature adaptive-modulation (RFaM) mechanism. RFaM can perceive the features that are semantically distinct enough in each semantic segmentation branch. Then, two pilot experiments based on the two branches are conducted to capture the significant prior semantic of two images, which then is applied to guide the fusion task in the integration of semantic segmentation branches and fusion branches. In addition, to aggregate both high-level semantics and impressive visual effects, we further investigate the frequency response of the prior semantics, and propose a multi-level representation-adaptive fusion (MRaF) module to explicitly integrate the low-frequent prior semantic with the high-frequent details. Extensive experiments on two public datasets demonstrate the superiority of our method over the state-of-the-art image fusion approaches, in terms of either the visual appeal or the high-level semantics.

Xu Wang, Jun Ma, Jing Li

Coronary CT angiography (CCTA) scans are widely used for diagnosis of coronary artery diseases. An accurate and automatic vessel labeling algorithm for CCTA analysis can significantly improve the diagnostic efficiency and reduce the clinicians'manual efforts. In this paper, we propose a simple vessel labeling method based on the Point Transformer, which only needs the coronary artery segmentation. Specifically, firstly, the coronary segmentation is transformed to point cloud. Then, these points are fed into the hierarchical transformer blocks to obtain the multi-level features, including local and global features. Finally, the network output the semantic classification points and map them to centerline labeling. This method is only based on the structure of coronary segmentation and need not other features, so it is easy to generalize to other vessel labeling tasks, e.g., head and neck vessel labeling. To evaluate the performance of our proposed method, CCTA scans of 53 subjects are collected in our experiment. The experimental results demonstrate the efficacy of this approach.

Beihang Song, Jing Li, Shan Xue et al.

Oriented object detection is a crucial task in computer vision. Current top-down oriented detection methods usually directly detect entire objects, and not only neglecting the authentic direction of targets, but also do not fully utilise the key semantic information, which causes a decrease in detection accuracy. In this study, we developed a single-stage rotating object detector via two points with a solar corona heatmap (ROTP) to detect oriented objects. The ROTP predicts parts of the object and then aggregates them to form a whole image. Herein, we meticulously represent an object in a random direction using the vertex, centre point with width, and height. Specifically, we regress two heatmaps that characterise the relative location of each object, which enhances the accuracy of locating objects and avoids deviations caused by angle predictions. To rectify the central misjudgement of the Gaussian heatmap on high-aspect ratio targets, we designed a solar corona heatmap generation method to improve the perception difference between the central and non-central samples. Additionally, we predicted the vertex relative to the direction of the centre point to connect two key points that belong to the same goal. Experiments on the HRSC 2016, UCASAOD, and DOTA datasets show that our ROTP achieves the most advanced performance with a simpler modelling and less manual intervention.

Yan Gao, Qimeng Wang, Xu Tang et al.

Non-maximum suppression (NMS) is widely used in object detection pipelines for removing duplicated bounding boxes. The inconsistency between the confidence for NMS and the real localization confidence seriously affects detection performance. Prior works propose to predict Intersection-over-Union (IoU) between bounding boxes and corresponding ground-truths to improve NMS, while accurately predicting IoU is still a challenging problem. We argue that the complex definition of IoU and feature misalignment make it difficult to predict IoU accurately. In this paper, we propose a novel Decoupled IoU Regression (DIR) model to handle these problems. The proposed DIR decouples the traditional localization confidence metric IoU into two new metrics, Purity and Integrity. Purity reflects the proportion of the object area in the detected bounding box, and Integrity refers to the completeness of the detected object area. Separately predicting Purity and Integrity can divide the complex mapping between the bounding box and its IoU into two clearer mappings and model them independently. In addition, a simple but effective feature realignment approach is also introduced to make the IoU regressor work in a hindsight manner, which can make the target mapping more stable. The proposed DIR can be conveniently integrated with existing two-stage detectors and significantly improve their performance. Through a simple implementation of DIR with HTC, we obtain 51.3% AP on MS COCO benchmark, which outperforms previous methods and achieves state-of-the-art.

Wuyang Chen, Xianzhi Du, Fan Yang et al.

This work presents a simple vision transformer design as a strong baseline for object localization and instance segmentation tasks. Transformers recently demonstrate competitive performance in image classification tasks. To adopt ViT to object detection and dense prediction tasks, many works inherit the multistage design from convolutional networks and highly customized ViT architectures. Behind this design, the goal is to pursue a better trade-off between computational cost and effective aggregation of multiscale global contexts. However, existing works adopt the multistage architectural design as a black-box solution without a clear understanding of its true benefits. In this paper, we comprehensively study three architecture design choices on ViT -- spatial reduction, doubled channels, and multiscale features -- and demonstrate that a vanilla ViT architecture can fulfill this goal without handcrafting multiscale features, maintaining the original ViT design philosophy. We further complete a scaling rule to optimize our model's trade-off on accuracy and computation cost / model size. By leveraging a constant feature resolution and hidden size throughout the encoder blocks, we propose a simple and compact ViT architecture called Universal Vision Transformer (UViT) that achieves strong performance on COCO object detection and instance segmentation tasks.

Li Jing, Jure Zbontar, Yann LeCun

An important component of autoencoders is the method by which the information capacity of the latent representation is minimized or limited. In this work, the rank of the covariance matrix of the codes is implicitly minimized by relying on the fact that gradient descent learning in multi-layer linear networks leads to minimum-rank solutions. By inserting a number of extra linear layers between the encoder and the decoder, the system spontaneously learns representations with a low effective dimension. The model, dubbed Implicit Rank-Minimizing Autoencoder (IRMAE), is simple, deterministic, and learns compact latent spaces. We demonstrate the validity of the method on several image generation and representation learning tasks.

Yang Liu, Zhaoyang Lu, Jing Li et al.

Recently, infrared human action recognition has attracted increasing attention for it has many advantages over visible light, that is, being robust to illumination change and shadows. However, the infrared action data is limited until now, which degrades the performance of infrared action recognition. Motivated by the idea of transfer learning, an infrared human action recognition framework using auxiliary data from visible light is proposed to solve the problem of limited infrared action data. In the proposed framework, we first construct a novel Cross-Dataset Feature Alignment and Generalization (CDFAG) framework to map the infrared data and visible light data into a common feature space, where Kernel Manifold Alignment (KEMA) and a dual alignedto-generalized encoders (AGE) model are employed to represent the feature. Then, a support vector machine (SVM) is trained, using both the infrared data and visible light data, and can classify the features derived from infrared data. The proposed method is evaluated on InfAR, which is a publicly available infrared human action dataset. To build up auxiliary data, we set up a novel visible light action dataset XD145. Experimental results show that the proposed method can achieve state-of-the-art performance compared with several transfer learning and domain adaptation methods.

Yang Liu, Zhaoyang Lu, Jing Li et al.

Infrared human action recognition has many advantages, i.e., it is insensitive to illumination change, appearance variability, and shadows. Existing methods for infrared action recognition are either based on spatial or local temporal information, however, the global temporal information, which can better describe the movements of body parts across the whole video, is not considered. In this letter, we propose a novel global temporal representation named optical-flow stacked difference image (OFSDI) and extract robust and discriminative feature from the infrared action data by considering the local, global, and spatial temporal information together. Due to the small size of the infrared action dataset, we first apply convolutional neural networks on local, spatial, and global temporal stream respectively to obtain efficient convolutional feature maps from the raw data rather than train a classifier directly. Then these convolutional feature maps are aggregated into effective descriptors named three-stream trajectory-pooled deep-convolutional descriptors by trajectory-constrained pooling. Furthermore, we improve the robustness of these features by using the locality-constrained linear coding (LLC) method. With these features, a linear support vector machine (SVM) is adopted to classify the action data in our scheme. We conduct the experiments on infrared action recognition datasets InfAR and NTU RGB+D. The experimental results show that the proposed approach outperforms the representative state-of-the-art handcrafted features and deep learning features based methods for the infrared action recognition.

Li Jing, Rumen Dangovski, Marin Soljacic

We present a logarithmic-scale efficient convolutional neural network architecture for edge devices, named WaveletNet. Our model is based on the well-known depthwise convolution, and on two new layers, which we introduce in this work: a wavelet convolution and a depthwise fast wavelet transform. By breaking the symmetry in channel dimensions and applying a fast algorithm, WaveletNet shrinks the complexity of convolutional blocks by an O(logD/D) factor, where D is the number of channels. Experiments on CIFAR-10 and ImageNet classification show superior and comparable performances of WaveletNet compared to state-of-the-art models such as MobileNetV2.

Yang Liu, Zhaoyang Lu, Jing Li et al.

Recognizing human actions from varied views is challenging due to huge appearance variations in different views. The key to this problem is to learn discriminant view-invariant representations generalizing well across views. In this paper, we address this problem by learning view-invariant representations hierarchically using a novel method, referred to as Joint Sparse Representation and Distribution Adaptation (JSRDA). To obtain robust and informative feature representations, we first incorporate a sample-affinity matrix into the marginalized stacked denoising Autoencoder (mSDA) to obtain shared features, which are then combined with the private features. In order to make the feature representations of videos across views transferable, we then learn a transferable dictionary pair simultaneously from pairs of videos taken at different views to encourage each action video across views to have the same sparse representation. However, the distribution difference across views still exists because a unified subspace where the sparse representations of one action across views are the same may not exist when the view difference is large. Therefore, we propose a novel unsupervised distribution adaptation method that learns a set of projections that project the source and target views data into respective low-dimensional subspaces where the marginal and conditional distribution differences are reduced simultaneously. Therefore, the finally learned feature representation is view-invariant and robust for substantial distribution difference across views even the view difference is large. Experimental results on four multiview datasets show that our approach outperforms the state-ofthe-art approaches.

Yichi Gu, Qisheng Wu, Jing Li et al.

Artificial intelligence is making great changes in academy and industry with the fast development of deep learning, which is a branch of machine learning and statistical learning. Fully convolutional network [1] is the standard model for semantic segmentation. Conditional random fields coded as CNN [2] or RNN [3] and connected with FCN has been successfully applied in object detection [4]. In this paper, we introduce a multi-resolution neural network for FCN and apply Gaussian filter to the extended CRF kernel neighborhood and the label image to reduce the oscillating effect of CRF neural network segmentation, thus achieve higher precision and faster training speed.