Yanan Sun

Aojun Lu, Tao Feng, Hangjie Yuan et al.

Efforts to overcome catastrophic forgetting have primarily centered around developing more effective Continual Learning (CL) methods. In contrast, less attention was devoted to analyzing the role of network architecture design (e.g., network depth, width, and components) in contributing to CL. This paper seeks to bridge this gap between network architecture design and CL, and to present a holistic study on the impact of network architectures on CL. This work considers architecture design at the network scaling level, i.e., width and depth, and also at the network components, i.e., skip connections, global pooling layers, and down-sampling. In both cases, we first derive insights through systematically exploring how architectural designs affect CL. Then, grounded in these insights, we craft a specialized search space for CL and further propose a simple yet effective ArchCraft method to steer a CL-friendly architecture, namely, this method recrafts AlexNet/ResNet into AlexAC/ResAC. Experimental validation across various CL settings and scenarios demonstrates that improved architectures are parameter-efficient, achieving state-of-the-art performance of CL while being 86%, 61%, and 97% more compact in terms of parameters than the naive CL architecture in Task IL and Class IL. Code is available at https://github.com/byyx666/ArchCraft.

Aojun Lu, Hangjie Yuan, Tao Feng et al.

The quest for Continual Learning (CL) seeks to empower neural networks with the ability to learn and adapt incrementally. Central to this pursuit is addressing the stability-plasticity dilemma, which involves striking a balance between two conflicting objectives: preserving previously learned knowledge and acquiring new knowledge. While numerous CL methods aim to achieve this trade-off, they often overlook the impact of network architecture on stability and plasticity, restricting the trade-off to the parameter level. In this paper, we delve into the conflict between stability and plasticity at the architectural level. We reveal that under an equal parameter constraint, deeper networks exhibit better plasticity, while wider networks are characterized by superior stability. To address this architectural-level dilemma, we introduce a novel framework denoted Dual-Arch, which serves as a plug-in component for CL. This framework leverages the complementary strengths of two distinct and independent networks: one dedicated to plasticity and the other to stability. Each network is designed with a specialized and lightweight architecture, tailored to its respective objective. Extensive experiments demonstrate that Dual-Arch enhances the performance of existing CL methods while being up to 87% more compact in terms of parameters. Code: https://github.com/byyx666/Dual-Arch.

Zetong Yang, Li Chen, Yanan Sun et al.

In contrast to extensive studies on general vision, pre-training for scalable visual autonomous driving remains seldom explored. Visual autonomous driving applications require features encompassing semantics, 3D geometry, and temporal information simultaneously for joint perception, prediction, and planning, posing dramatic challenges for pre-training. To resolve this, we bring up a new pre-training task termed as visual point cloud forecasting - predicting future point clouds from historical visual input. The key merit of this task captures the synergic learning of semantics, 3D structures, and temporal dynamics. Hence it shows superiority in various downstream tasks. To cope with this new problem, we present ViDAR, a general model to pre-train downstream visual encoders. It first extracts historical embeddings by the encoder. These representations are then transformed to 3D geometric space via a novel Latent Rendering operator for future point cloud prediction. Experiments show significant gain in downstream tasks, e.g., 3.1% NDS on 3D detection, ~10% error reduction on motion forecasting, and ~15% less collision rate on planning.

Yanchen Liu, Yanan Sun, Zhening Xing et al.

Existing text-to-video methods struggle to transfer motion smoothly from a reference object to a target object with significant differences in appearance or structure between them. To address this challenge, we introduce MotionShot, a training-free framework capable of parsing reference-target correspondences in a fine-grained manner, thereby achieving high-fidelity motion transfer while preserving coherence in appearance. To be specific, MotionShot first performs semantic feature matching to ensure high-level alignments between the reference and target objects. It then further establishes low-level morphological alignments through reference-to-target shape retargeting. By encoding motion with temporal attention, our MotionShot can coherently transfer motion across objects, even in the presence of significant appearance and structure disparities, demonstrated by extensive experiments. The project page is available at: https://motionshot.github.io/.

Yuwei Ou, Yuqi Feng, Yanan Sun

To defend deep neural networks from adversarial attacks, adversarial training has been drawing increasing attention for its effectiveness. However, the accuracy and robustness resulting from the adversarial training are limited by the architecture, because adversarial training improves accuracy and robustness by adjusting the weight connection affiliated to the architecture. In this work, we propose ARNAS to search for accurate and robust architectures for adversarial training. First we design an accurate and robust search space, in which the placement of the cells and the proportional relationship of the filter numbers are carefully determined. With the design, the architectures can obtain both accuracy and robustness by deploying accurate and robust structures to their sensitive positions, respectively. Then we propose a differentiable multi-objective search strategy, performing gradient descent towards directions that are beneficial for both natural loss and adversarial loss, thus the accuracy and robustness can be guaranteed at the same time. We conduct comprehensive experiments in terms of white-box attacks, black-box attacks, and transferability. Experimental results show that the searched architecture has the strongest robustness with the competitive accuracy, and breaks the traditional idea that NAS-based architectures cannot transfer well to complex tasks in robustness scenarios. By analyzing outstanding architectures searched, we also conclude that accurate and robust neural architectures tend to deploy different structures near the input and output, which has great practical significance on both hand-crafting and automatically designing of accurate and robust architectures.

Yuqiao Liu, Yehui Tang, Yanan Sun

Neural Architecture Search (NAS) can automatically design well-performed architectures of Deep Neural Networks (DNNs) for the tasks at hand. However, one bottleneck of NAS is the prohibitively computational cost largely due to the expensive performance evaluation. The neural predictors can directly estimate the performance without any training of the DNNs to be evaluated, thus have drawn increasing attention from researchers. Despite their popularity, they also suffer a severe limitation: the shortage of annotated DNN architectures for effectively training the neural predictors. In this paper, we proposed Homogeneous Architecture Augmentation for Neural Predictor (HAAP) of DNN architectures to address the issue aforementioned. Specifically, a homogeneous architecture augmentation algorithm is proposed in HAAP to generate sufficient training data taking the use of homogeneous representation. Furthermore, the one-hot encoding strategy is introduced into HAAP to make the representation of DNN architectures more effective. The experiments have been conducted on both NAS-Benchmark-101 and NAS-Bench-201 dataset. The experimental results demonstrate that the proposed HAAP algorithm outperforms the state of the arts compared, yet with much less training data. In addition, the ablation studies on both benchmark datasets have also shown the universality of the homogeneous architecture augmentation.

Jindi Lv, Qing Ye, Yanan Sun et al.

Arrhythmia is a cardiovascular disease that manifests irregular heartbeats. In arrhythmia detection, the electrocardiogram (ECG) signal is an important diagnostic technique. However, manually evaluating ECG signals is a complicated and time-consuming task. With the application of convolutional neural networks (CNNs), the evaluation process has been accelerated and the performance is improved. It is noteworthy that the performance of CNNs heavily depends on their architecture design, which is a complex process grounded on expert experience and trial-and-error. In this paper, we propose a novel approach, Heart-Darts, to efficiently classify the ECG signals by automatically designing the CNN model with the differentiable architecture search (i.e., Darts, a cell-based neural architecture search method). Specifically, we initially search a cell architecture by Darts and then customize a novel CNN model for ECG classification based on the obtained cells. To investigate the efficiency of the proposed method, we evaluate the constructed model on the MIT-BIH arrhythmia database. Additionally, the extensibility of the proposed CNN model is validated on two other new databases. Extensive experimental results demonstrate that the proposed method outperforms several state-of-the-art CNN models in ECG classification in terms of both performance and generalization capability.

Yanan Sun, Guanzhi Wang, Qiao Gu et al.

Despite the significant progress made by deep learning in natural image matting, there has been so far no representative work on deep learning for video matting due to the inherent technical challenges in reasoning temporal domain and lack of large-scale video matting datasets. In this paper, we propose a deep learning-based video matting framework which employs a novel and effective spatio-temporal feature aggregation module (ST-FAM). As optical flow estimation can be very unreliable within matting regions, ST-FAM is designed to effectively align and aggregate information across different spatial scales and temporal frames within the network decoder. To eliminate frame-by-frame trimap annotations, a lightweight interactive trimap propagation network is also introduced. The other contribution consists of a large-scale video matting dataset with groundtruth alpha mattes for quantitative evaluation and real-world high-resolution videos with trimaps for qualitative evaluation. Quantitative and qualitative experimental results show that our framework significantly outperforms conventional video matting and deep image matting methods applied to video in presence of multi-frame temporal information.

Yanan Sun, Chi-Keung Tang, Yu-Wing Tai

Natural image matting separates the foreground from background in fractional occupancy which can be caused by highly transparent objects, complex foreground (e.g., net or tree), and/or objects containing very fine details (e.g., hairs). Although conventional matting formulation can be applied to all of the above cases, no previous work has attempted to reason the underlying causes of matting due to various foreground semantics. We show how to obtain better alpha mattes by incorporating into our framework semantic classification of matting regions. Specifically, we consider and learn 20 classes of matting patterns, and propose to extend the conventional trimap to semantic trimap. The proposed semantic trimap can be obtained automatically through patch structure analysis within trimap regions. Meanwhile, we learn a multi-class discriminator to regularize the alpha prediction at semantic level, and content-sensitive weights to balance different regularization losses. Experiments on multiple benchmarks show that our method outperforms other methods and has achieved the most competitive state-of-the-art performance. Finally, we contribute a large-scale Semantic Image Matting Dataset with careful consideration of data balancing across different semantic classes.

Qing Ye, Yuxuan Han, Yanan sun et al.

Parameter updating is an important stage in parallelism-based distributed deep learning. Synchronous methods are widely used in distributed training the Deep Neural Networks (DNNs). To reduce the communication and synchronization overhead of synchronous methods, decreasing the synchronization frequency (e.g., every $n$ mini-batches) is a straightforward approach. However, it often suffers from poor convergence. In this paper, we propose a new algorithm of integrating Particle Swarm Optimization (PSO) into the distributed training process of DNNs to automatically compute new parameters. In the proposed algorithm, a computing work is encoded by a particle, the weights of DNNs and the training loss are modeled by the particle attributes. At each synchronization stage, the weights are updated by PSO from the sub weights gathered from all workers, instead of averaging the weights or the gradients. To verify the performance of the proposed algorithm, the experiments are performed on two commonly used image classification benchmarks: MNIST and CIFAR10, and compared with the peer competitors at multiple different synchronization configurations. The experimental results demonstrate the competitiveness of the proposed algorithm.

Yanan Sun, Xian Sun, Yuhan Fang et al.

Evolutionary Neural Architecture Search (ENAS) can automatically design the architectures of Deep Neural Networks (DNNs) using evolutionary computation algorithms. However, most ENAS algorithms require intensive computational resource, which is not necessarily available to the users interested. Performance predictors are a type of regression models which can assist to accomplish the search, while without exerting much computational resource. Despite various performance predictors have been designed, they employ the same training protocol to build the regression models: 1) sampling a set of DNNs with performance as the training dataset, 2) training the model with the mean square error criterion, and 3) predicting the performance of DNNs newly generated during the ENAS. In this paper, we point out that the three steps constituting the training protocol are not well though-out through intuitive and illustrative examples. Furthermore, we propose a new training protocol to address these issues, consisting of designing a pairwise ranking indicator to construct the training target, proposing to use the logistic regression to fit the training samples, and developing a differential method to building the training instances. To verify the effectiveness of the proposed training protocol, four widely used regression models in the field of machine learning have been chosen to perform the comparisons on two benchmark datasets. The experimental results of all the comparisons demonstrate that the proposed training protocol can significantly improve the performance prediction accuracy against the traditional training protocols.

Yuqiao Liu, Yanan Sun, Bing Xue et al.

Deep Neural Networks (DNNs) have achieved great success in many applications. The architectures of DNNs play a crucial role in their performance, which is usually manually designed with rich expertise. However, such a design process is labour intensive because of the trial-and-error process, and also not easy to realize due to the rare expertise in practice. Neural Architecture Search (NAS) is a type of technology that can design the architectures automatically. Among different methods to realize NAS, Evolutionary Computation (EC) methods have recently gained much attention and success. Unfortunately, there has not yet been a comprehensive summary of the EC-based NAS algorithms. This paper reviews over 200 papers of most recent EC-based NAS methods in light of the core components, to systematically discuss their design principles as well as justifications on the design. Furthermore, current challenges and issues are also discussed to identify future research in this emerging field.

Yuqiao Liu, Yanan Sun, Bing Xue et al.

Hyperspectral images (HSIs) are susceptible to various noise factors leading to the loss of information, and the noise restricts the subsequent HSIs object detection and classification tasks. In recent years, learning-based methods have demonstrated their superior strengths in denoising the HSIs. Unfortunately, most of the methods are manually designed based on the extensive expertise that is not necessarily available to the users interested. In this paper, we propose a novel algorithm to automatically build an optimal Convolutional Neural Network (CNN) to effectively denoise HSIs. Particularly, the proposed algorithm focuses on the architectures and the initialization of the connection weights of the CNN. The experiments of the proposed algorithm have been well-designed and compared against the state-of-the-art peer competitors, and the experimental results demonstrate the competitive performance of the proposed algorithm in terms of the different evaluation metrics, visual assessments, and the computational complexity.

Lei Ke, Shichao Li, Yanan Sun et al.

We present a novel end-to-end framework named as GSNet (Geometric and Scene-aware Network), which jointly estimates 6DoF poses and reconstructs detailed 3D car shapes from single urban street view. GSNet utilizes a unique four-way feature extraction and fusion scheme and directly regresses 6DoF poses and shapes in a single forward pass. Extensive experiments show that our diverse feature extraction and fusion scheme can greatly improve model performance. Based on a divide-and-conquer 3D shape representation strategy, GSNet reconstructs 3D vehicle shape with great detail (1352 vertices and 2700 faces). This dense mesh representation further leads us to consider geometrical consistency and scene context, and inspires a new multi-objective loss function to regularize network training, which in turn improves the accuracy of 6D pose estimation and validates the merit of jointly performing both tasks. We evaluate GSNet on the largest multi-task ApolloCar3D benchmark and achieve state-of-the-art performance both quantitatively and qualitatively. Project page is available at https://lkeab.github.io/gsnet/.

Qing Ye, Yuhao Zhou, Mingjia Shi et al.

Synchronous strategies with data parallelism, such as the Synchronous StochasticGradient Descent (S-SGD) and the model averaging methods, are widely utilizedin distributed training of Deep Neural Networks (DNNs), largely owing to itseasy implementation yet promising performance. Particularly, each worker ofthe cluster hosts a copy of the DNN and an evenly divided share of the datasetwith the fixed mini-batch size, to keep the training of DNNs convergence. In thestrategies, the workers with different computational capability, need to wait foreach other because of the synchronization and delays in network transmission,which will inevitably result in the high-performance workers wasting computation.Consequently, the utilization of the cluster is relatively low. To alleviate thisissue, we propose the Dynamic Batch Size (DBS) strategy for the distributedtraining of DNNs. Specifically, the performance of each worker is evaluatedfirst based on the fact in the previous epoch, and then the batch size and datasetpartition are dynamically adjusted in consideration of the current performanceof the worker, thereby improving the utilization of the cluster. To verify theeffectiveness of the proposed strategy, extensive experiments have been conducted,and the experimental results indicate that the proposed strategy can fully utilizethe performance of the cluster, reduce the training time, and have good robustnesswith disturbance by irrelevant tasks. Furthermore, rigorous theoretical analysis hasalso been provided to prove the convergence of the proposed strategy.

Zetong Yang, Yanan Sun, Shu Liu et al.

Currently, there have been many kinds of voxel-based 3D single stage detectors, while point-based single stage methods are still underexplored. In this paper, we first present a lightweight and effective point-based 3D single stage object detector, named 3DSSD, achieving a good balance between accuracy and efficiency. In this paradigm, all upsampling layers and refinement stage, which are indispensable in all existing point-based methods, are abandoned to reduce the large computation cost. We novelly propose a fusion sampling strategy in downsampling process to make detection on less representative points feasible. A delicate box prediction network including a candidate generation layer, an anchor-free regression head with a 3D center-ness assignment strategy is designed to meet with our demand of accuracy and speed. Our paradigm is an elegant single stage anchor-free framework, showing great superiority to other existing methods. We evaluate 3DSSD on widely used KITTI dataset and more challenging nuScenes dataset. Our method outperforms all state-of-the-art voxel-based single stage methods by a large margin, and has comparable performance to two stage point-based methods as well, with inference speed more than 25 FPS, 2x faster than former state-of-the-art point-based methods.

Yanan Sun, Ziyao Ren, Gary G. Yen et al.

The superiority of Convolutional Neural Networks (CNNs) largely relies on their architectures that are often manually crafted with extensive human expertise. Unfortunately, such kind of domain knowledge is not necessarily owned by each of the users interested. Data mining on existing CNN can discover useful patterns and fundamental sub-comments from their architectures, providing researchers with strong prior knowledge to design proper CNN architectures when they have no expertise in CNNs. There have been various state-of-the-art data mining algorithms at hand, while there is only rare work that has been done for the mining. One of the main reasons is the gap between CNN architectures and data mining algorithms. Specifically, the current CNN architecture descriptions cannot be exactly vectorized to the input of data mining algorithms. In this paper, we propose a unified approach, named ArcText, to describing CNN architectures based on text. Particularly, four different units and an ordering method have been elaborately designed in ArcText, to uniquely describe the same architecture with sufficient information. Also, the resulted description can be exactly converted back to the corresponding CNN architecture. ArcText bridges the gap between CNN architectures and data mining researchers, and has the potentiality to be utilized to wider scenarios.

Zetong Yang, Yanan Sun, Shu Liu et al.

We present a new two-stage 3D object detection framework, named sparse-to-dense 3D Object Detector (STD). The first stage is a bottom-up proposal generation network that uses raw point cloud as input to generate accurate proposals by seeding each point with a new spherical anchor. It achieves a high recall with less computation compared with prior works. Then, PointsPool is applied for generating proposal features by transforming their interior point features from sparse expression to compact representation, which saves even more computation time. In box prediction, which is the second stage, we implement a parallel intersection-over-union (IoU) branch to increase awareness of localization accuracy, resulting in further improved performance. We conduct experiments on KITTI dataset, and evaluate our method in terms of 3D object and Bird's Eye View (BEV) detection. Our method outperforms other state-of-the-arts by a large margin, especially on the hard set, with inference speed more than 10 FPS.

Bin Wang, Yanan Sun, Bing Xue et al.

In recent years, convolutional neural networks (CNNs) have become deeper in order to achieve better classification accuracy in image classification. However, it is difficult to deploy the state-of-the-art deep CNNs for industrial use due to the difficulty of manually fine-tuning the hyperparameters and the trade-off between classification accuracy and computational cost. This paper proposes a novel multi-objective optimization method for evolving state-of-the-art deep CNNs in real-life applications, which automatically evolves the non-dominant solutions at the Pareto front. Three major contributions are made: Firstly, a new encoding strategy is designed to encode one of the best state-of-the-art CNNs; With the classification accuracy and the number of floating point operations as the two objectives, a multi-objective particle swarm optimization method is developed to evolve the non-dominant solutions; Last but not least, a new infrastructure is designed to boost the experiments by concurrently running the experiments on multiple GPUs across multiple machines, and a Python library is developed and released to manage the infrastructure. The experimental results demonstrate that the non-dominant solutions found by the proposed algorithm form a clear Pareto front, and the proposed infrastructure is able to almost linearly reduce the running time.

Zetong Yang, Yanan Sun, Shu Liu et al.

We present a novel 3D object detection framework, named IPOD, based on raw point cloud. It seeds object proposal for each point, which is the basic element. This paradigm provides us with high recall and high fidelity of information, leading to a suitable way to process point cloud data. We design an end-to-end trainable architecture, where features of all points within a proposal are extracted from the backbone network and achieve a proposal feature for final bounding inference. These features with both context information and precise point cloud coordinates yield improved performance. We conduct experiments on KITTI dataset, evaluating our performance in terms of 3D object detection, Bird's Eye View (BEV) detection and 2D object detection. Our method accomplishes new state-of-the-art , showing great advantage on the hard set.

Yanan Sun, Bing Xue, Mengjie Zhang et al.

The performance of Convolutional Neural Networks (CNNs) highly relies on their architectures. In order to design a CNN with promising performance, extended expertise in both CNNs and the investigated problem is required, which is not necessarily held by every user interested in CNNs or the problem domain. In this paper, we propose to automatically evolve CNN architectures by using a genetic algorithm based on ResNet blocks and DenseNet blocks. The proposed algorithm is \textbf{completely} automatic in designing CNN architectures, particularly, neither pre-processing before it starts nor post-processing on the designed CNN is needed. Furthermore, the proposed algorithm does not require users with domain knowledge on CNNs, the investigated problem or even genetic algorithms. The proposed algorithm is evaluated on CIFAR10 and CIFAR100 against 18 state-of-the-art peer competitors. Experimental results show that it outperforms state-of-the-art CNNs hand-crafted and CNNs designed by automatic peer competitors in terms of the classification accuracy, and achieves the competitive classification accuracy against semi-automatic peer competitors. In addition, the proposed algorithm consumes much less time than most peer competitors in finding the best CNN architectures.

Yanan Sun, Bing Xue, Mengjie Zhang et al.

Convolutional Neural Networks (CNNs) have gained a remarkable success on many image classification tasks in recent years. However, the performance of CNNs highly relies upon their architectures. For most state-of-the-art CNNs, their architectures are often manually-designed with expertise in both CNNs and the investigated problems. Therefore, it is difficult for users, who have no extended expertise in CNNs, to design optimal CNN architectures for their own image classification problems of interest. In this paper, we propose an automatic CNN architecture design method by using genetic algorithms, to effectively address the image classification tasks. The most merit of the proposed algorithm remains in its "automatic" characteristic that users do not need domain knowledge of CNNs when using the proposed algorithm, while they can still obtain a promising CNN architecture for the given images. The proposed algorithm is validated on widely used benchmark image classification datasets, by comparing to the state-of-the-art peer competitors covering eight manually-designed CNNs, seven automatic+manually tuning and five automatic CNN architecture design algorithms. The experimental results indicate the proposed algorithm outperforms the existing automatic CNN architecture design algorithms in terms of classification accuracy, parameter numbers and consumed computational resources. The proposed algorithm also shows the very comparable classification accuracy to the best one from manually-designed and automatic+manually tuning CNNs, while consumes much less of computational resource.

Yanan Sun, Bing Xue, Mengjie Zhang et al.

Convolutional auto-encoders have shown their remarkable performance in stacking to deep convolutional neural networks for classifying image data during past several years. However, they are unable to construct the state-of-the-art convolutional neural networks due to their intrinsic architectures. In this regard, we propose a flexible convolutional auto-encoder by eliminating the constraints on the numbers of convolutional layers and pooling layers from the traditional convolutional auto-encoder. We also design an architecture discovery method by using particle swarm optimization, which is capable of automatically searching for the optimal architectures of the proposed flexible convolutional auto-encoder with much less computational resource and without any manual intervention. We use the designed architecture optimization algorithm to test the proposed flexible convolutional auto-encoder through utilizing one graphic processing unit card on four extensively used image classification datasets. Experimental results show that our work in this paper significantly outperform the peer competitors including the state-of-the-art algorithm.

Yanan Sun, Bing Xue, Mengjie Zhang et al.

Evolutionary computation methods have been successfully applied to neural networks since two decades ago, while those methods cannot scale well to the modern deep neural networks due to the complicated architectures and large quantities of connection weights. In this paper, we propose a new method using genetic algorithms for evolving the architectures and connection weight initialization values of a deep convolutional neural network to address image classification problems. In the proposed algorithm, an efficient variable-length gene encoding strategy is designed to represent the different building blocks and the unpredictable optimal depth in convolutional neural networks. In addition, a new representation scheme is developed for effectively initializing connection weights of deep convolutional neural networks, which is expected to avoid networks getting stuck into local minima which is typically a major issue in the backward gradient-based optimization. Furthermore, a novel fitness evaluation method is proposed to speed up the heuristic search with substantially less computational resource. The proposed algorithm is examined and compared with 22 existing algorithms on nine widely used image classification tasks, including the state-of-the-art methods. The experimental results demonstrate the remarkable superiority of the proposed algorithm over the state-of-the-art algorithms in terms of classification error rate and the number of parameters (weights).