Ruofeng Tong

Huimin Huang, Shiao Xie1, Lanfen Lin et al.

Recently, a variety of vision transformers have been developed as their capability of modeling long-range dependency. In current transformer-based backbones for medical image segmentation, convolutional layers were replaced with pure transformers, or transformers were added to the deepest encoder to learn global context. However, there are mainly two challenges in a scale-wise perspective: (1) intra-scale problem: the existing methods lacked in extracting local-global cues in each scale, which may impact the signal propagation of small objects; (2) inter-scale problem: the existing methods failed to explore distinctive information from multiple scales, which may hinder the representation learning from objects with widely variable size, shape and location. To address these limitations, we propose a novel backbone, namely ScaleFormer, with two appealing designs: (1) A scale-wise intra-scale transformer is designed to couple the CNN-based local features with the transformer-based global cues in each scale, where the row-wise and column-wise global dependencies can be extracted by a lightweight Dual-Axis MSA. (2) A simple and effective spatial-aware inter-scale transformer is designed to interact among consensual regions in multiple scales, which can highlight the cross-scale dependency and resolve the complex scale variations. Experimental results on different benchmarks demonstrate that our Scale-Former outperforms the current state-of-the-art methods. The code is publicly available at: https://github.com/ZJUGiveLab/ScaleFormer.

Hongyi Wang, Luyang Luo, Fang Wang et al.

Whole Slide Image (WSI) classification remains a challenge due to their extremely high resolution and the absence of fine-grained labels. Presently, WSI classification is usually regarded as a Multiple Instance Learning (MIL) problem when only slide-level labels are available. MIL methods involve a patch embedding module and a bag-level classification module, but they are prohibitively expensive to be trained in an end-to-end manner. Therefore, existing methods usually train them separately, or directly skip the training of the embedder. Such schemes hinder the patch embedder's access to slide-level semantic labels, resulting in inconsistency within the entire MIL pipeline. To overcome this issue, we propose a novel framework called Iteratively Coupled MIL (ICMIL), which bridges the loss back-propagation process from the bag-level classifier to the patch embedder. In ICMIL, we use category information in the bag-level classifier to guide the patch-level fine-tuning of the patch feature extractor. The refined embedder then generates better instance representations for achieving a more accurate bag-level classifier. By coupling the patch embedder and bag classifier at a low cost, our proposed framework enables information exchange between the two modules, benefiting the entire MIL classification model. We tested our framework on two datasets using three different backbones, and our experimental results demonstrate consistent performance improvements over state-of-the-art MIL methods. The code is available at: https://github.com/Dootmaan/ICMIL.

Dan Song, Xuanpu Zhang, Juan Zhou et al.

Image-based virtual try-on aims to synthesize a naturally dressed person image with a clothing image, which revolutionizes online shopping and inspires related topics within image generation, showing both research significance and commercial potential. However, there is a gap between current research progress and commercial applications and an absence of comprehensive overview of this field to accelerate the development.In this survey, we provide a comprehensive analysis of the state-of-the-art techniques and methodologies in aspects of pipeline architecture, person representation and key modules such as try-on indication, clothing warping and try-on stage. We additionally apply CLIP to assess the semantic alignment of try-on results, and evaluate representative methods with uniformly implemented evaluation metrics on the same dataset.In addition to quantitative and qualitative evaluation of current open-source methods, unresolved issues are highlighted and future research directions are prospected to identify key trends and inspire further exploration. The uniformly implemented evaluation metrics, dataset and collected methods will be made public available at https://github.com/little-misfit/Survey-Of-Virtual-Try-On.

Li Ye, Xinhang Zhou, Xingyu Yang et al.

Mid-surface abstraction is essential for finite element analysis of thin-walled CAD models. Existing face pairing-based methods rely on handcrafted geometric heuristics, yet real-world industrial models frequently exhibit multi-wall-thickness regions, self-matching face configurations, and demand for non-center offset surfaces--scenarios where rule-based approaches consistently fail. We present MidSurfNet, a learning-augmented framework that addresses these limitations through two novel components: (1) a neural face pairing module that learns to predict face pair confidence from geometric and topological features, handling complex pairing scenarios beyond rule-based methods; and (2) an interference implicit field that represents mid-surfaces as the interference of two signed distance functions, enabling generalized offset control for flexible positioning in downstream CAE/FEA-oriented workflows. We construct a large-scale mid-surface dataset containing over 1,500 manually annotated CAD models. Experiments demonstrate that MidSurfNet achieves 87.32% face pairing accuracy and successfully handles multi-wall-thickness (61.90% completion) and self-matching (52.94% completion) scenarios that confound all existing methods. Furthermore, MidSurfNet provides a learning-based approach to generalized mid-surface abstraction with arbitrary offset control for CAE-oriented applications.

Shuyi Ouyang, Hongyi Wang, Ziwei Niu et al.

The task of multi-label image classification involves recognizing multiple objects within a single image. Considering both valuable semantic information contained in the labels and essential visual features presented in the image, tight visual-linguistic interactions play a vital role in improving classification performance. Moreover, given the potential variance in object size and appearance within a single image, attention to features of different scales can help to discover possible objects in the image. Recently, Transformer-based methods have achieved great success in multi-label image classification by leveraging the advantage of modeling long-range dependencies, but they have several limitations. Firstly, existing methods treat visual feature extraction and cross-modal fusion as separate steps, resulting in insufficient visual-linguistic alignment in the joint semantic space. Additionally, they only extract visual features and perform cross-modal fusion at a single scale, neglecting objects with different characteristics. To address these issues, we propose a Hierarchical Scale-Aware Vision-Language Transformer (HSVLT) with two appealing designs: (1)~A hierarchical multi-scale architecture that involves a Cross-Scale Aggregation module, which leverages joint multi-modal features extracted from multiple scales to recognize objects of varying sizes and appearances in images. (2)~Interactive Visual-Linguistic Attention, a novel attention mechanism module that tightly integrates cross-modal interaction, enabling the joint updating of visual, linguistic and multi-modal features. We have evaluated our method on three benchmark datasets. The experimental results demonstrate that HSVLT surpasses state-of-the-art methods with lower computational cost.

Hongyi Wang, Lanfen Lin, Hongjie Hu et al.

High resolution (HR) 3D images are widely used nowadays, such as medical images like Magnetic Resonance Imaging (MRI) and Computed Tomography (CT). However, segmentation of these 3D images remains a challenge due to their high spatial resolution and dimensionality in contrast to currently limited GPU memory. Therefore, most existing 3D image segmentation methods use patch-based models, which have low inference efficiency and ignore global contextual information. To address these problems, we propose a super-resolution (SR) based patch-free 3D image segmentation framework that can realize HR segmentation from a global-wise low-resolution (LR) input. The framework contains two sub-tasks, of which semantic segmentation is the main task and super resolution is an auxiliary task aiding in rebuilding the high frequency information from the LR input. To furthermore balance the information loss with the LR input, we propose a High-Frequency Guidance Module (HGM), and design an efficient selective cropping algorithm to crop an HR patch from the original image as restoration guidance for it. In addition, we also propose a Task-Fusion Module (TFM) to exploit the inter connections between segmentation and SR task, realizing joint optimization of the two tasks. When predicting, only the main segmentation task is needed, while other modules can be removed for acceleration. The experimental results on two different datasets show that our framework has a four times higher inference speed compared to traditional patch-based methods, while its performance also surpasses other patch-based and patch-free models.

Jiahua Dong, Yue Zhang, Qiuli Wang et al.

Medical image segmentation is crucial in the field of medical imaging, aiding in disease diagnosis and surgical planning. Most established segmentation methods rely on supervised deep learning, in which clean and precise labels are essential for supervision and significantly impact the performance of models. However, manually delineated labels often contain noise, such as missing labels and inaccurate boundary delineation, which can hinder networks from correctly modeling target characteristics. In this paper, we propose a deep self-cleansing segmentation framework that can preserve clean labels while cleansing noisy ones in the training phase. To achieve this, we devise a gaussian mixture model-based label filtering module that distinguishes noisy labels from clean labels. Additionally, we develop a label cleansing module to generate pseudo low-noise labels for identified noisy samples. The preserved clean labels and pseudo-labels are then used jointly to supervise the network. Validated on a clinical liver tumor dataset and a public cardiac diagnosis dataset, our method can effectively suppress the interference from noisy labels and achieve prominent segmentation performance.

Jiahua Dong, Guohua Cheng, Yue Zhang et al.

Multi-organ segmentation, which identifies and separates different organs in medical images, is a fundamental task in medical image analysis. Recently, the immense success of deep learning motivated its wide adoption in multi-organ segmentation tasks. However, due to expensive labor costs and expertise, the availability of multi-organ annotations is usually limited and hence poses a challenge in obtaining sufficient training data for deep learning-based methods. In this paper, we aim to address this issue by combining off-the-shelf single-organ segmentation models to develop a multi-organ segmentation model on the target dataset, which helps get rid of the dependence on annotated data for multi-organ segmentation. To this end, we propose a novel dual-stage method that consists of a Model Adaptation stage and a Model Ensemble stage. The first stage enhances the generalization of each off-the-shelf segmentation model on the target domain, while the second stage distills and integrates knowledge from multiple adapted single-organ segmentation models. Extensive experiments on four abdomen datasets demonstrate that our proposed method can effectively leverage off-the-shelf single-organ segmentation models to obtain a tailored model for multi-organ segmentation with high accuracy.

Yinao Li, Yutaro Iwamoto, Ryousuke Nakamura et al.

Recently, convolutional neural networks (CNN) have obtained promising results in single-image SR for hyperspectral pansharpening. However, enhancing CNNs' representation ability with fewer parameters and a shorter prediction time is a challenging and critical task. In this paper, we propose a novel multi-spectral image fusion method using a combination of the previously proposed 3D CNN model VolumeNet and 2.5D texture transfer method using other modality high resolution (HR) images. Since a multi-spectral (MS) image consists of several bands and each band is a 2D image slice, MS images can be seen as 3D data. Thus, we use the previously proposed VolumeNet to fuse HR panchromatic (PAN) images and bicubic interpolated MS images. Because the proposed 3D VolumeNet can effectively improve the accuracy by expanding the receptive field of the model, and due to its lightweight structure, we can achieve better performance against the existing method without purchasing a large number of remote sensing images for training. In addition, VolumeNet can restore the high-frequency information lost in the HR MR image as much as possible, reducing the difficulty of feature extraction in the following step: 2.5D texture transfer. As one of the latest technologies, deep learning-based texture transfer has been demonstrated to effectively and efficiently improve the visual performance and quality evaluation indicators of image reconstruction. Different from the texture transfer processing of RGB image, we use HR PAN images as the reference images and perform texture transfer for each frequency band of MS images, which is named 2.5D texture transfer. The experimental results show that the proposed method outperforms the existing methods in terms of objective accuracy assessment, method efficiency, and visual subjective evaluation.

Hongyi Wang, Shiao Xie, Lanfen Lin et al.

Though U-Net has achieved tremendous success in medical image segmentation tasks, it lacks the ability to explicitly model long-range dependencies. Therefore, Vision Transformers have emerged as alternative segmentation structures recently, for their innate ability of capturing long-range correlations through Self-Attention (SA). However, Transformers usually rely on large-scale pre-training and have high computational complexity. Furthermore, SA can only model self-affinities within a single sample, ignoring the potential correlations of the overall dataset. To address these problems, we propose a novel Transformer module named Mixed Transformer Module (MTM) for simultaneous inter- and intra- affinities learning. MTM first calculates self-affinities efficiently through our well-designed Local-Global Gaussian-Weighted Self-Attention (LGG-SA). Then, it mines inter-connections between data samples through External Attention (EA). By using MTM, we construct a U-shaped model named Mixed Transformer U-Net (MT-UNet) for accurate medical image segmentation. We test our method on two different public datasets, and the experimental results show that the proposed method achieves better performance over other state-of-the-art methods. The code is available at: https://github.com/Dootmaan/MT-UNet.

Huimin Huang, Lanfen Lin, Ruofeng Tong et al.

Recently, a growing interest has been seen in deep learning-based semantic segmentation. UNet, which is one of deep learning networks with an encoder-decoder architecture, is widely used in medical image segmentation. Combining multi-scale features is one of important factors for accurate segmentation. UNet++ was developed as a modified Unet by designing an architecture with nested and dense skip connections. However, it does not explore sufficient information from full scales and there is still a large room for improvement. In this paper, we propose a novel UNet 3+, which takes advantage of full-scale skip connections and deep supervisions. The full-scale skip connections incorporate low-level details with high-level semantics from feature maps in different scales; while the deep supervision learns hierarchical representations from the full-scale aggregated feature maps. The proposed method is especially benefiting for organs that appear at varying scales. In addition to accuracy improvements, the proposed UNet 3+ can reduce the network parameters to improve the computation efficiency. We further propose a hybrid loss function and devise a classification-guided module to enhance the organ boundary and reduce the over-segmentation in a non-organ image, yielding more accurate segmentation results. The effectiveness of the proposed method is demonstrated on two datasets. The code is available at: github.com/ZJUGiveLab/UNet-Version

Xiaolong Yin, Xingyu Lu, Jiahang Shen et al.

A CAD command sequence is a typical parametric design paradigm in 3D CAD systems where a model is constructed by overlaying 2D sketches with operations such as extrusion, revolution, and Boolean operations. Although there is growing academic interest in the automatic generation of command sequences, existing methods and datasets only support operations such as 2D sketching, extrusion,and Boolean operations. This limitation makes it challenging to represent more complex geometries. In this paper, we present a reinforcement learning (RL) training environment (gym) built on a CAD geometric engine. Given an input boundary representation (B-Rep) geometry, the policy network in the RL algorithm generates an action. This action, along with previously generated actions, is processed within the gym to produce the corresponding CAD geometry, which is then fed back into the policy network. The rewards, determined by the difference between the generated and target geometries within the gym, are used to update the RL network. Our method supports operations beyond sketches, Boolean, and extrusion, including revolution operations. With this training gym, we achieve state-of-the-art (SOTA) quality in generating command sequences from B-Rep geometries.

Jingzhe Ni, Xiaolong Yin, Xingyu Lu et al.

Computer-Aided Design (CAD) plays a pivotal role in industrial manufacturing but typically requires a high level of expertise from designers. To lower the entry barrier and improve design efficiency, we present an agent for CAD conceptual design powered by large language models (LLMs). The agent accepts both abstract textual descriptions and freehand sketches as input, engaging in interactive dialogue with users to refine and clarify design requirements through comprehensive requirement analysis. Built upon a novel Context-Independent Imperative Paradigm (CIP), the agent generates high-quality CAD modeling code. During the generation process, the agent incorporates iterative visual feedback to improve model quality. Generated design cases are stored in a structured knowledge base, enabling continuous improvement of the agent's code generation capabilities. Experimental results demonstrate that our method achieves state-of-the-art performance in CAD code generation.

Yudi Li, Min Tang, Yun Yang et al.

We present a novel learning method to predict the cloth deformation for skeleton-based characters with a two-stream network. The characters processed in our approach are not limited to humans, and can be other skeletal-based representations of non-human targets such as fish or pets. We use a novel network architecture which consists of skeleton-based and mesh-based residual networks to learn the coarse and wrinkle features as the overall residual from the template cloth mesh. Our network is used to predict the deformation for loose or tight-fitting clothing or dresses. We ensure that the memory footprint of our network is low, and thereby result in reduced storage and computational requirements. In practice, our prediction for a single cloth mesh for the skeleton-based character takes about 7 milliseconds on an NVIDIA GeForce RTX 3090 GPU. Compared with prior methods, our network can generate fine deformation results with details and wrinkles.

Xu Ma, Junkun Yuan, Yen-wei Chen et al.

Unsupervised domain adaptation (UDA) aims to learn transferable knowledge from a labeled source domain and adapts a trained model to an unlabeled target domain. To bridge the gap between source and target domains, one prevailing strategy is to minimize the distribution discrepancy by aligning their semantic features extracted by deep models. The existing alignment-based methods mainly focus on reducing domain divergence in the same model layer. However, the same level of semantic information could distribute across model layers due to the domain shifts. To further boost model adaptation performance, we propose a novel method called Attention-based Cross-layer Domain Alignment (ACDA), which captures the semantic relationship between the source and target domains across model layers and calibrates each level of semantic information automatically through a dynamic attention mechanism. An elaborate attention mechanism is designed to reweight each cross-layer pair based on their semantic similarity for precise domain alignment, effectively matching each level of semantic information during model adaptation. Extensive experiments on multiple benchmark datasets consistently show that the proposed method ACDA yields state-of-the-art performance.

Yudi Li, Min Tang, Yun Yang et al.

We present a novel mesh-based learning approach (N-Cloth) for plausible 3D cloth deformation prediction. Our approach is general and can handle cloth or obstacles represented by triangle meshes with arbitrary topologies. We use graph convolution to transform the cloth and object meshes into a latent space to reduce the non-linearity in the mesh space. Our network can predict the target 3D cloth mesh deformation based on the initial state of the cloth mesh template and the target obstacle mesh. Our approach can handle complex cloth meshes with up to 100K triangles and scenes with various objects corresponding to SMPL humans, non-SMPL humans or rigid bodies. In practice, our approach can be used to generate plausible cloth simulation at 30-45 fps on an NVIDIA GeForce RTX 3090 GPU. We highlight its benefits over prior learning-based methods and physically-based cloth simulators.

Diqiong Jiang, Yiwei Jin, Fanglue Zhang et al.

3D Morphable Models (3DMMs) are generative models for face shape and appearance. However, the shape parameters of traditional 3DMMs satisfy the multivariate Gaussian distribution while the identity embeddings satisfy the hypersphere distribution, and this conflict makes it challenging for face reconstruction models to preserve the faithfulness and the shape consistency simultaneously. To address this issue, we propose the Sphere Face Model(SFM), a novel 3DMM for monocular face reconstruction, which can preserve both shape fidelity and identity consistency. The core of our SFM is the basis matrix which can be used to reconstruct 3D face shapes, and the basic matrix is learned by adopting a two-stage training approach where 3D and 2D training data are used in the first and second stages, respectively. To resolve the distribution mismatch, we design a novel loss to make the shape parameters have a hyperspherical latent space. Extensive experiments show that SFM has high representation ability and shape parameter space's clustering performance. Moreover, it produces fidelity face shapes, and the shapes are consistent in challenging conditions in monocular face reconstruction.

Yue Zhang, Chengtao Peng, Liying Peng et al.

Multi-phase computed tomography (CT) images provide crucial complementary information for accurate liver tumor segmentation (LiTS). State-of-the-art multi-phase LiTS methods usually fused cross-phase features through phase-weighted summation or channel-attention based concatenation. However, these methods ignored the spatial (pixel-wise) relationships between different phases, hence leading to insufficient feature integration. In addition, the performance of existing methods remains subject to the uncertainty in segmentation, which is particularly acute in tumor boundary regions. In this work, we propose a novel LiTS method to adequately aggregate multi-phase information and refine uncertain region segmentation. To this end, we introduce a spatial aggregation module (SAM), which encourages per-pixel interactions between different phases, to make full use of cross-phase information. Moreover, we devise an uncertain region inpainting module (URIM) to refine uncertain pixels using neighboring discriminative features. Experiments on an in-house multi-phase CT dataset of focal liver lesions (MPCT-FLLs) demonstrate that our method achieves promising liver tumor segmentation and outperforms state-of-the-arts.

Diqiong Jiang, Yiwei Jin, Fanglue Zhang et al.

Many recent works have reconstructed distinctive 3D face shapes by aggregating shape parameters of the same identity and separating those of different people based on parametric models (e.g., 3D morphable models (3DMMs)). However, despite the high accuracy in the face recognition task using these shape parameters, the visual discrimination of face shapes reconstructed from those parameters is unsatisfactory. The following research question has not been answered in previous works: Do discriminative shape parameters guarantee visual discrimination in represented 3D face shapes? This paper analyzes the relationship between shape parameters and reconstructed shape geometry and proposes a novel shape identity-aware regularization(SIR) loss for shape parameters, aiming at increasing discriminability in both the shape parameter and shape geometry domains. Moreover, to cope with the lack of training data containing both landmark and identity annotations, we propose a network structure and an associated training strategy to leverage mixed data containing either identity or landmark labels. We compare our method with existing methods in terms of the reconstruction error, visual distinguishability, and face recognition accuracy of the shape parameters. Experimental results show that our method outperforms the state-of-the-art methods.

Huimin Huang, Ming Cai, Lanfen Lin et al.

Coronavirus Disease 2019 (COVID-19) has rapidly spread in 2020, emerging a mass of studies for lung infection segmentation from CT images. Though many methods have been proposed for this issue, it is a challenging task because of infections of various size appearing in different lobe zones. To tackle these issues, we propose a Graph-based Pyramid Global Context Reasoning (Graph-PGCR) module, which is capable of modeling long-range dependencies among disjoint infections as well as adapt size variation. We first incorporate graph convolution to exploit long-term contextual information from multiple lobe zones. Different from previous average pooling or maximum object probability, we propose a saliency-aware projection mechanism to pick up infection-related pixels as a set of graph nodes. After graph reasoning, the relation-aware features are reversed back to the original coordinate space for the down-stream tasks. We further construct multiple graphs with different sampling rates to handle the size variation problem. To this end, distinct multi-scale long-range contextual patterns can be captured. Our Graph-PGCR module is plug-and-play, which can be integrated into any architecture to improve its performance. Experiments demonstrated that the proposed method consistently boost the performance of state-of-the-art backbone architectures on both of public and our private COVID-19 datasets.

Yingying Xu, Ming Cai, Lanfen Lin et al.

In this paper, we propose a phase attention residual network (PA-ResSeg) to model multi-phase features for accurate liver tumor segmentation, in which a phase attention (PA) is newly proposed to additionally exploit the images of arterial (ART) phase to facilitate the segmentation of portal venous (PV) phase. The PA block consists of an intra-phase attention (Intra-PA) module and an inter-phase attention (Inter-PA) module to capture channel-wise self-dependencies and cross-phase interdependencies, respectively. Thus it enables the network to learn more representative multi-phase features by refining the PV features according to the channel dependencies and recalibrating the ART features based on the learned interdependencies between phases. We propose a PA-based multi-scale fusion (MSF) architecture to embed the PA blocks in the network at multiple levels along the encoding path to fuse multi-scale features from multi-phase images. Moreover, a 3D boundary-enhanced loss (BE-loss) is proposed for training to make the network more sensitive to boundaries. To evaluate the performance of our proposed PA-ResSeg, we conducted experiments on a multi-phase CT dataset of focal liver lesions (MPCT-FLLs). Experimental results show the effectiveness of the proposed method by achieving a dice per case (DPC) of 0.77.87, a dice global (DG) of 0.8682, a volumetric overlap error (VOE) of 0.3328 and a relative volume difference (RVD) of 0.0443 on the MPCT-FLLs. Furthermore, to validate the effectiveness and robustness of PA-ResSeg, we conducted extra experiments on another multi-phase liver tumor dataset and obtained a DPC of 0.8290, a DG of 0.9132, a VOE of 0.2637 and a RVD of 0.0163. The proposed method shows its robustness and generalization capability in different datasets and different backbones.

Titinunt Kitrungrotsakul, Iwamoto Yutaro, Lanfen Lin et al.

Deep learning techniques have successfully been employed in numerous computer vision tasks including image segmentation. The techniques have also been applied to medical image segmentation, one of the most critical tasks in computer-aided diagnosis. Compared with natural images, the medical image is a gray-scale image with low-contrast (even with some invisible parts). Because some organs have similar intensity and texture with neighboring organs, there is usually a need to refine automatic segmentation results. In this paper, we propose an interactive deep refinement framework to improve the traditional semantic segmentation networks such as U-Net and fully convolutional network. In the proposed framework, we added a refinement network to traditional segmentation network to refine the segmentation results.Experimental results with public dataset revealed that the proposed method could achieve higher accuracy than other state-of-the-art methods.

Chengyang Li, Dan Song, Ruofeng Tong et al.

Multispectral pedestrian detection has attracted increasing attention from the research community due to its crucial competence for many around-the-clock applications (e.g., video surveillance and autonomous driving), especially under insufficient illumination conditions. We create a human baseline over the KAIST dataset and reveal that there is still a large gap between current top detectors and human performance. To narrow this gap, we propose a network fusion architecture, which consists of a multispectral proposal network to generate pedestrian proposals, and a subsequent multispectral classification network to distinguish pedestrian instances from hard negatives. The unified network is learned by jointly optimizing pedestrian detection and semantic segmentation tasks. The final detections are obtained by integrating the outputs from different modalities as well as the two stages. The approach significantly outperforms state-of-the-art methods on the KAIST dataset while remain fast. Additionally, we contribute a sanitized version of training annotations for the KAIST dataset, and examine the effects caused by different kinds of annotation errors. Future research of this problem will benefit from the sanitized version which eliminates the interference of annotation errors.

Chengyang Li, Dan Song, Ruofeng Tong et al.

Multispectral images of color-thermal pairs have shown more effective than a single color channel for pedestrian detection, especially under challenging illumination conditions. However, there is still a lack of studies on how to fuse the two modalities effectively. In this paper, we deeply compare six different convolutional network fusion architectures and analyse their adaptations, enabling a vanilla architecture to obtain detection performances comparable to the state-of-the-art results. Further, we discover that pedestrian detection confidences from color or thermal images are correlated with illumination conditions. With this in mind, we propose an Illumination-aware Faster R-CNN (IAF RCNN). Specifically, an Illumination-aware Network is introduced to give an illumination measure of the input image. Then we adaptively merge color and thermal sub-networks via a gate function defined over the illumination value. The experimental results on KAIST Multispectral Pedestrian Benchmark validate the effectiveness of the proposed IAF R-CNN.