Wei Li

Pan Zhang, Xiaoyi Dong, Bin Wang et al. · pku

We propose InternLM-XComposer, a vision-language large model that enables advanced image-text comprehension and composition. The innovative nature of our model is highlighted by three appealing properties: 1) Interleaved Text-Image Composition: InternLM-XComposer can effortlessly generate coherent and contextual articles that seamlessly integrate images, providing a more engaging and immersive reading experience. Simply provide a writing instruction, and our system will generate the corresponding manuscript. It can intelligently identify the areas in the text where images would enhance the content and automatically insert the most appropriate visual candidates. 2) Comprehension with Rich Multilingual Knowledge: The text-image comprehension is empowered by training on an extensive multi-modal multilingual database with carefully crafted strategies, resulting in a deep understanding of visual content. 3) State-of-the-art Performance: Our model consistently achieves state-of-the-art results across various mainstream benchmarks for vision-language foundational models, including MME Benchmark, MMBench, MMBench-CN, Seed-Bench, CCBench (Chinese Cultural Benchmark), QBench and Tiny LVLM. Owing to the absence of established metrics for quantitatively assessing text-image composition, we have devised a robust evaluation procedure that comprises both human and GPT4-Vision (GPT4-V) to ensure reliability. Notably, our InternLM-XComposer achieves competitive text-image composition scores compared to public solutions, including GPT4-V and GPT3.5. Collectively, InternLM-XComposer seamlessly blends advanced text-image comprehension and composition, revolutionizing vision-language interaction and offering new insights and opportunities. The InternLM-XComposer model series are publicly available at https://github.com/InternLM/InternLM-XComposer.

Jiashi Li, Xin Xia, Wei Li et al.

Due to the complex attention mechanisms and model design, most existing vision Transformers (ViTs) can not perform as efficiently as convolutional neural networks (CNNs) in realistic industrial deployment scenarios, e.g. TensorRT and CoreML. This poses a distinct challenge: Can a visual neural network be designed to infer as fast as CNNs and perform as powerful as ViTs? Recent works have tried to design CNN-Transformer hybrid architectures to address this issue, yet the overall performance of these works is far away from satisfactory. To end these, we propose a next generation vision Transformer for efficient deployment in realistic industrial scenarios, namely Next-ViT, which dominates both CNNs and ViTs from the perspective of latency/accuracy trade-off. In this work, the Next Convolution Block (NCB) and Next Transformer Block (NTB) are respectively developed to capture local and global information with deployment-friendly mechanisms. Then, Next Hybrid Strategy (NHS) is designed to stack NCB and NTB in an efficient hybrid paradigm, which boosts performance in various downstream tasks. Extensive experiments show that Next-ViT significantly outperforms existing CNNs, ViTs and CNN-Transformer hybrid architectures with respect to the latency/accuracy trade-off across various vision tasks. On TensorRT, Next-ViT surpasses ResNet by 5.5 mAP (from 40.4 to 45.9) on COCO detection and 7.7% mIoU (from 38.8% to 46.5%) on ADE20K segmentation under similar latency. Meanwhile, it achieves comparable performance with CSWin, while the inference speed is accelerated by 3.6x. On CoreML, Next-ViT surpasses EfficientFormer by 4.6 mAP (from 42.6 to 47.2) on COCO detection and 3.5% mIoU (from 45.1% to 48.6%) on ADE20K segmentation under similar latency. Our code and models are made public at: https://github.com/bytedance/Next-ViT

Yan Wang, Junbo Yin, Wei Li et al.

LiDAR-based 3D object detection is an indispensable task in advanced autonomous driving systems. Though impressive detection results have been achieved by superior 3D detectors, they suffer from significant performance degeneration when facing unseen domains, such as different LiDAR configurations, different cities, and weather conditions. The mainstream approaches tend to solve these challenges by leveraging unsupervised domain adaptation (UDA) techniques. However, these UDA solutions just yield unsatisfactory 3D detection results when there is a severe domain shift, e.g., from Waymo (64-beam) to nuScenes (32-beam). To address this, we present a novel Semi-Supervised Domain Adaptation method for 3D object detection (SSDA3D), where only a few labeled target data is available, yet can significantly improve the adaptation performance. In particular, our SSDA3D includes an Inter-domain Adaptation stage and an Intra-domain Generalization stage. In the first stage, an Inter-domain Point-CutMix module is presented to efficiently align the point cloud distribution across domains. The Point-CutMix generates mixed samples of an intermediate domain, thus encouraging to learn domain-invariant knowledge. Then, in the second stage, we further enhance the model for better generalization on the unlabeled target set. This is achieved by exploring Intra-domain Point-MixUp in semi-supervised learning, which essentially regularizes the pseudo label distribution. Experiments from Waymo to nuScenes show that, with only 10% labeled target data, our SSDA3D can surpass the fully-supervised oracle model with 100% target label. Our code is available at https://github.com/yinjunbo/SSDA3D.

Wenhao Wang, Yifan Sun, Wei Li et al.

This paper explores a hierarchical prompting mechanism for the hierarchical image classification (HIC) task. Different from prior HIC methods, our hierarchical prompting is the first to explicitly inject ancestor-class information as a tokenized hint that benefits the descendant-class discrimination. We think it well imitates human visual recognition, i.e., humans may use the ancestor class as a prompt to draw focus on the subtle differences among descendant classes. We model this prompting mechanism into a Transformer with Hierarchical Prompting (TransHP). TransHP consists of three steps: 1) learning a set of prompt tokens to represent the coarse (ancestor) classes, 2) on-the-fly predicting the coarse class of the input image at an intermediate block, and 3) injecting the prompt token of the predicted coarse class into the intermediate feature. Though the parameters of TransHP maintain the same for all input images, the injected coarse-class prompt conditions (modifies) the subsequent feature extraction and encourages a dynamic focus on relatively subtle differences among the descendant classes. Extensive experiments show that TransHP improves image classification on accuracy (e.g., improving ViT-B/16 by +2.83% ImageNet classification accuracy), training data efficiency (e.g., +12.69% improvement under 10% ImageNet training data), and model explainability. Moreover, TransHP also performs favorably against prior HIC methods, showing that TransHP well exploits the hierarchical information. The code is available at: https://github.com/WangWenhao0716/TransHP.

Zhiyang Chen, Yousong Zhu, Zhaowen Li et al.

Visual tasks vary a lot in their output formats and concerned contents, therefore it is hard to process them with an identical structure. One main obstacle lies in the high-dimensional outputs in object-level visual tasks. In this paper, we propose an object-centric vision framework, Obj2Seq. Obj2Seq takes objects as basic units, and regards most object-level visual tasks as sequence generation problems of objects. Therefore, these visual tasks can be decoupled into two steps. First recognize objects of given categories, and then generate a sequence for each of these objects. The definition of the output sequences varies for different tasks, and the model is supervised by matching these sequences with ground-truth targets. Obj2Seq is able to flexibly determine input categories to satisfy customized requirements, and be easily extended to different visual tasks. When experimenting on MS COCO, Obj2Seq achieves 45.7% AP on object detection, 89.0% AP on multi-label classification and 65.0% AP on human pose estimation. These results demonstrate its potential to be generally applied to different visual tasks. Code has been made available at: https://github.com/CASIA-IVA-Lab/Obj2Seq.

Yuhang Zang, Wei Li, Kaiyang Zhou et al.

Prompt tuning, a parameter- and data-efficient transfer learning paradigm that tunes only a small number of parameters in a model's input space, has become a trend in the vision community since the emergence of large vision-language models like CLIP. We present a systematic study on two representative prompt tuning methods, namely text prompt tuning and visual prompt tuning. A major finding is that none of the unimodal prompt tuning methods performs consistently well: text prompt tuning fails on data with high intra-class visual variances while visual prompt tuning cannot handle low inter-class variances. To combine the best from both worlds, we propose a simple approach called Unified Prompt Tuning (UPT), which essentially learns a tiny neural network to jointly optimize prompts across different modalities. Extensive experiments on over 11 vision datasets show that UPT achieves a better trade-off than the unimodal counterparts on few-shot learning benchmarks, as well as on domain generalization benchmarks. Code and models will be released to facilitate future research.

Anthony Cioppa, Silvio Giancola, Vladimir Somers et al. · pku

The SoccerNet 2023 challenges were the third annual video understanding challenges organized by the SoccerNet team. For this third edition, the challenges were composed of seven vision-based tasks split into three main themes. The first theme, broadcast video understanding, is composed of three high-level tasks related to describing events occurring in the video broadcasts: (1) action spotting, focusing on retrieving all timestamps related to global actions in soccer, (2) ball action spotting, focusing on retrieving all timestamps related to the soccer ball change of state, and (3) dense video captioning, focusing on describing the broadcast with natural language and anchored timestamps. The second theme, field understanding, relates to the single task of (4) camera calibration, focusing on retrieving the intrinsic and extrinsic camera parameters from images. The third and last theme, player understanding, is composed of three low-level tasks related to extracting information about the players: (5) re-identification, focusing on retrieving the same players across multiple views, (6) multiple object tracking, focusing on tracking players and the ball through unedited video streams, and (7) jersey number recognition, focusing on recognizing the jersey number of players from tracklets. Compared to the previous editions of the SoccerNet challenges, tasks (2-3-7) are novel, including new annotations and data, task (4) was enhanced with more data and annotations, and task (6) now focuses on end-to-end approaches. More information on the tasks, challenges, and leaderboards are available on https://www.soccer-net.org. Baselines and development kits can be found on https://github.com/SoccerNet.

Jiahao Xie, Wei Li, Xiaohang Zhan et al.

We present Masked Frequency Modeling (MFM), a unified frequency-domain-based approach for self-supervised pre-training of visual models. Instead of randomly inserting mask tokens to the input embeddings in the spatial domain, in this paper, we shift the perspective to the frequency domain. Specifically, MFM first masks out a portion of frequency components of the input image and then predicts the missing frequencies on the frequency spectrum. Our key insight is that predicting masked components in the frequency domain is more ideal to reveal underlying image patterns rather than predicting masked patches in the spatial domain, due to the heavy spatial redundancy. Our findings suggest that with the right configuration of mask-and-predict strategy, both the structural information within high-frequency components and the low-level statistics among low-frequency counterparts are useful in learning good representations. For the first time, MFM demonstrates that, for both ViT and CNN, a simple non-Siamese framework can learn meaningful representations even using none of the following: (i) extra data, (ii) extra model, (iii) mask token. Experimental results on image classification and semantic segmentation, as well as several robustness benchmarks show the competitive performance and advanced robustness of MFM compared with recent masked image modeling approaches. Furthermore, we also comprehensively investigate the effectiveness of classical image restoration tasks for representation learning from a unified frequency perspective and reveal their intriguing relations with our MFM approach.

Heng-Chao Li, Wen-Shuai Hu, Wei Li et al.

The problem of effectively exploiting the information multiple data sources has become a relevant but challenging research topic in remote sensing. In this paper, we propose a new approach to exploit the complementarity of two data sources: hyperspectral images (HSIs) and light detection and ranging (LiDAR) data. Specifically, we develop a new dual-channel spatial, spectral and multiscale attention convolutional long short-term memory neural network (called dual-channel A3CLNN) for feature extraction and classification of multisource remote sensing data. Spatial, spectral and multiscale attention mechanisms are first designed for HSI and LiDAR data in order to learn spectral- and spatial-enhanced feature representations, and to represent multiscale information for different classes. In the designed fusion network, a novel composite attention learning mechanism (combined with a three-level fusion strategy) is used to fully integrate the features in these two data sources. Finally, inspired by the idea of transfer learning, a novel stepwise training strategy is designed to yield a final classification result. Our experimental results, conducted on several multisource remote sensing data sets, demonstrate that the newly proposed dual-channel A3CLNN exhibits better feature representation ability (leading to more competitive classification performance) than other state-of-the-art methods.

Yuxiang Zhang, Wei Li, Weidong Sun et al.

Currently, cross-scene hyperspectral image (HSI) classification has drawn increasing attention. It is necessary to train a model only on source domain (SD) and directly transferring the model to target domain (TD), when TD needs to be processed in real time and cannot be reused for training. Based on the idea of domain generalization, a Single-source Domain Expansion Network (SDEnet) is developed to ensure the reliability and effectiveness of domain extension. The method uses generative adversarial learning to train in SD and test in TD. A generator including semantic encoder and morph encoder is designed to generate the extended domain (ED) based on encoder-randomization-decoder architecture, where spatial and spectral randomization are specifically used to generate variable spatial and spectral information, and the morphological knowledge is implicitly applied as domain invariant information during domain expansion. Furthermore, the supervised contrastive learning is employed in the discriminator to learn class-wise domain invariant representation, which drives intra-class samples of SD and ED. Meanwhile, adversarial training is designed to optimize the generator to drive intra-class samples of SD and ED to be separated. Extensive experiments on two public HSI datasets and one additional multispectral image (MSI) dataset demonstrate the superiority of the proposed method when compared with state-of-the-art techniques.

Eduardo Pérez-Pellitero, Sibi Catley-Chandar, Richard Shaw et al.

This paper reviews the challenge on constrained high dynamic range (HDR) imaging that was part of the New Trends in Image Restoration and Enhancement (NTIRE) workshop, held in conjunction with CVPR 2022. This manuscript focuses on the competition set-up, datasets, the proposed methods and their results. The challenge aims at estimating an HDR image from multiple respective low dynamic range (LDR) observations, which might suffer from under- or over-exposed regions and different sources of noise. The challenge is composed of two tracks with an emphasis on fidelity and complexity constraints: In Track 1, participants are asked to optimize objective fidelity scores while imposing a low-complexity constraint (i.e. solutions can not exceed a given number of operations). In Track 2, participants are asked to minimize the complexity of their solutions while imposing a constraint on fidelity scores (i.e. solutions are required to obtain a higher fidelity score than the prescribed baseline). Both tracks use the same data and metrics: Fidelity is measured by means of PSNR with respect to a ground-truth HDR image (computed both directly and with a canonical tonemapping operation), while complexity metrics include the number of Multiply-Accumulate (MAC) operations and runtime (in seconds).

Zhanchao Huang, Wei Li, Xiang-Gen Xia et al.

A few lightweight convolutional neural network (CNN) models have been recently designed for remote sensing object detection (RSOD). However, most of them simply replace vanilla convolutions with stacked separable convolutions, which may not be efficient due to a lot of precision losses and may not be able to detect oriented bounding boxes (OBB). Also, the existing OBB detection methods are difficult to constrain the shape of objects predicted by CNNs accurately. In this paper, we propose an effective lightweight oriented object detector (LO-Det). Specifically, a channel separation-aggregation (CSA) structure is designed to simplify the complexity of stacked separable convolutions, and a dynamic receptive field (DRF) mechanism is developed to maintain high accuracy by customizing the convolution kernel and its perception range dynamically when reducing the network complexity. The CSA-DRF component optimizes efficiency while maintaining high accuracy. Then, a diagonal support constraint head (DSC-Head) component is designed to detect OBBs and constrain their shapes more accurately and stably. Extensive experiments on public datasets demonstrate that the proposed LO-Det can run very fast even on embedded devices with the competitive accuracy of detecting oriented objects.

Yuchao Wang, Haochen Wang, Yujun Shen et al.

The crux of semi-supervised semantic segmentation is to assign adequate pseudo-labels to the pixels of unlabeled images. A common practice is to select the highly confident predictions as the pseudo ground-truth, but it leads to a problem that most pixels may be left unused due to their unreliability. We argue that every pixel matters to the model training, even its prediction is ambiguous. Intuitively, an unreliable prediction may get confused among the top classes (i.e., those with the highest probabilities), however, it should be confident about the pixel not belonging to the remaining classes. Hence, such a pixel can be convincingly treated as a negative sample to those most unlikely categories. Based on this insight, we develop an effective pipeline to make sufficient use of unlabeled data. Concretely, we separate reliable and unreliable pixels via the entropy of predictions, push each unreliable pixel to a category-wise queue that consists of negative samples, and manage to train the model with all candidate pixels. Considering the training evolution, where the prediction becomes more and more accurate, we adaptively adjust the threshold for the reliable-unreliable partition. Experimental results on various benchmarks and training settings demonstrate the superiority of our approach over the state-of-the-art alternatives.

Silvio Giancola, Anthony Cioppa, Adrien Deliège et al.

The SoccerNet 2022 challenges were the second annual video understanding challenges organized by the SoccerNet team. In 2022, the challenges were composed of 6 vision-based tasks: (1) action spotting, focusing on retrieving action timestamps in long untrimmed videos, (2) replay grounding, focusing on retrieving the live moment of an action shown in a replay, (3) pitch localization, focusing on detecting line and goal part elements, (4) camera calibration, dedicated to retrieving the intrinsic and extrinsic camera parameters, (5) player re-identification, focusing on retrieving the same players across multiple views, and (6) multiple object tracking, focusing on tracking players and the ball through unedited video streams. Compared to last year's challenges, tasks (1-2) had their evaluation metrics redefined to consider tighter temporal accuracies, and tasks (3-6) were novel, including their underlying data and annotations. More information on the tasks, challenges and leaderboards are available on https://www.soccer-net.org. Baselines and development kits are available on https://github.com/SoccerNet.

Yong Cao, Wei Li, Xianzhi Li et al.

Sign language recognition and translation first uses a recognition module to generate glosses from sign language videos and then employs a translation module to translate glosses into spoken sentences. Most existing works focus on the recognition step, while paying less attention to sign language translation. In this work, we propose a task-aware instruction network, namely TIN-SLT, for sign language translation, by introducing the instruction module and the learning-based feature fuse strategy into a Transformer network. In this way, the pre-trained model's language ability can be well explored and utilized to further boost the translation performance. Moreover, by exploring the representation space of sign language glosses and target spoken language, we propose a multi-level data augmentation scheme to adjust the data distribution of the training set. We conduct extensive experiments on two challenging benchmark datasets, PHOENIX-2014-T and ASLG-PC12, on which our method outperforms former best solutions by 1.65 and 1.42 in terms of BLEU-4. Our code is published at https://github.com/yongcaoplus/TIN-SLT.

Wei Li, Jiahao Xie, Chen Change Loy

We introduce Correlational Image Modeling (CIM), a novel and surprisingly effective approach to self-supervised visual pre-training. Our CIM performs a simple pretext task: we randomly crop image regions (exemplars) from an input image (context) and predict correlation maps between the exemplars and the context. Three key designs enable correlational image modeling as a nontrivial and meaningful self-supervisory task. First, to generate useful exemplar-context pairs, we consider cropping image regions with various scales, shapes, rotations, and transformations. Second, we employ a bootstrap learning framework that involves online and target encoders. During pre-training, the former takes exemplars as inputs while the latter converts the context. Third, we model the output correlation maps via a simple cross-attention block, within which the context serves as queries and the exemplars offer values and keys. We show that CIM performs on par or better than the current state of the art on self-supervised and transfer benchmarks.

Xuming Zhang, Jian Yan, Jia Tian et al.

Deep learning methods have been successfully applied to hyperspectral image (HSI) classification with remarkable performance. Because of limited labelled HSI data, earlier studies primarily adopted a patch-based classification framework, which divides images into overlapping patches for training and testing. However, this approach results in redundant computations and possible information leakage. In this study, we propose an objective evaluation-based high-efficiency learning framework for tiny HSI classification. This framework comprises two main parts: (i) a leakage-free balanced sampling strategy, and (ii) a modified end-to-end fully convolutional network (FCN) architecture that optimizes the trade-off between accuracy and efficiency. The leakage-free balanced sampling strategy generates balanced and non-overlapping training and testing data by partitioning an HSI and the ground truth image into small windows, each of which corresponds to one training or testing sample. The proposed high-efficiency FCN exhibits a pixel-to-pixel architecture with modifications aimed at faster inference speed and improved parameter efficiency. Experiments conducted on four representative datasets demonstrated that the proposed sampling strategy can provide objective performance evaluation and that the proposed network outperformed many state-of-the-art approaches with respect to the speed/accuracy tradeoff. Our source code is available at https://github.com/xmzhang2018.

Chenzhong Gao, Wei Li, Ran Tao et al.

Multi-source image registration is challenging due to intensity, rotation, and scale differences among the images. Considering the characteristics and differences of multi-source remote sensing images, a feature-based registration algorithm named Multi-scale Histogram of Local Main Orientation (MS-HLMO) is proposed. Harris corner detection is first adopted to generate feature points. The HLMO feature of each Harris feature point is extracted on a Partial Main Orientation Map (PMOM) with a Generalized Gradient Location and Orientation Histogram-like (GGLOH) feature descriptor, which provides high intensity, rotation, and scale invariance. The feature points are matched through a multi-scale matching strategy. Comprehensive experiments on 17 multi-source remote sensing scenes demonstrate that the proposed MS-HLMO and its simplified version MS-HLMO$^+$ outperform other competitive registration algorithms in terms of effectiveness and generalization.

Zhaowen Li, Yousong Zhu, Fan Yang et al.

Self-supervised learning (SSL) holds promise in leveraging large amounts of unlabeled data. However, the success of popular SSL methods has limited on single-centric-object images like those in ImageNet and ignores the correlation among the scene and instances, as well as the semantic difference of instances in the scene. To address the above problems, we propose a Unified Self-supervised Visual Pre-training (UniVIP), a novel self-supervised framework to learn versatile visual representations on either single-centric-object or non-iconic dataset. The framework takes into account the representation learning at three levels: 1) the similarity of scene-scene, 2) the correlation of scene-instance, 3) the discrimination of instance-instance. During the learning, we adopt the optimal transport algorithm to automatically measure the discrimination of instances. Massive experiments show that UniVIP pre-trained on non-iconic COCO achieves state-of-the-art transfer performance on a variety of downstream tasks, such as image classification, semi-supervised learning, object detection and segmentation. Furthermore, our method can also exploit single-centric-object dataset such as ImageNet and outperforms BYOL by 2.5% with the same pre-training epochs in linear probing, and surpass current self-supervised object detection methods on COCO dataset, demonstrating its universality and potential.

Wenxuan Ma, Jinming Zhang, Shuang Li et al. · tsinghua

Extensive studies on Unsupervised Domain Adaptation (UDA) have propelled the deployment of deep learning from limited experimental datasets into real-world unconstrained domains. Most UDA approaches align features within a common embedding space and apply a shared classifier for target prediction. However, since a perfectly aligned feature space may not exist when the domain discrepancy is large, these methods suffer from two limitations. First, the coercive domain alignment deteriorates target domain discriminability due to lacking target label supervision. Second, the source-supervised classifier is inevitably biased to source data, thus it may underperform in target domain. To alleviate these issues, we propose to simultaneously conduct feature alignment in two individual spaces focusing on different domains, and create for each space a domain-oriented classifier tailored specifically for that domain. Specifically, we design a Domain-Oriented Transformer (DOT) that has two individual classification tokens to learn different domain-oriented representations, and two classifiers to preserve domain-wise discriminability. Theoretical guaranteed contrastive-based alignment and the source-guided pseudo-label refinement strategy are utilized to explore both domain-invariant and specific information. Comprehensive experiments validate that our method achieves state-of-the-art on several benchmarks.

Hai Wu, Chenglu Wen, Wei Li et al.

3D object detection received increasing attention in autonomous driving recently. Objects in 3D scenes are distributed with diverse orientations. Ordinary detectors do not explicitly model the variations of rotation and reflection transformations. Consequently, large networks and extensive data augmentation are required for robust detection. Recent equivariant networks explicitly model the transformation variations by applying shared networks on multiple transformed point clouds, showing great potential in object geometry modeling. However, it is difficult to apply such networks to 3D object detection in autonomous driving due to its large computation cost and slow reasoning speed. In this work, we present TED, an efficient Transformation-Equivariant 3D Detector to overcome the computation cost and speed issues. TED first applies a sparse convolution backbone to extract multi-channel transformation-equivariant voxel features; and then aligns and aggregates these equivariant features into lightweight and compact representations for high-performance 3D object detection. On the highly competitive KITTI 3D car detection leaderboard, TED ranked 1st among all submissions with competitive efficiency.

Ao Wang, Wei Li, Xin Wu et al.

Infrared small target detection (ISTD) has attracted widespread attention and been applied in various fields. Due to the small size of infrared targets and the noise interference from complex backgrounds, the performance of ISTD using convolutional neural networks (CNNs) is restricted. Moreover, the constriant that long-distance dependent features can not be encoded by the vanilla CNNs also impairs the robustness of capturing targets' shapes and locations in complex scenarios. To this end, a multi-patch attention network (MPANet) based on the axial-attention encoder and the multi-scale patch branch (MSPB) structure is proposed. Specially, an axial-attention-improved encoder architecture is designed to highlight the effective features of small targets and suppress background noises. Furthermore, the developed MSPB structure fuses the coarse-grained and fine-grained features from different semantic scales. Extensive experiments on the SIRST dataset show the superiority performance and effectiveness of the proposed MPANet compared to the state-of-the-art methods.

Yuchao Wang, Jingjing Fei, Haochen Wang et al.

Semantic segmentation usually suffers from a long-tail data distribution. Due to the imbalanced number of samples across categories, the features of those tail classes may get squeezed into a narrow area in the feature space. Towards a balanced feature distribution, we introduce category-wise variation into the network predictions in the training phase such that an instance is no longer projected to a feature point, but a small region instead. Such a perturbation is highly dependent on the category scale, which appears as assigning smaller variation to head classes and larger variation to tail classes. In this way, we manage to close the gap between the feature areas of different categories, resulting in a more balanced representation. It is noteworthy that the introduced variation is discarded at the inference stage to facilitate a confident prediction. Although with an embarrassingly simple implementation, our method manifests itself in strong generalizability to various datasets and task settings. Extensive experiments suggest that our plug-in design lends itself well to a range of state-of-the-art approaches and boosts the performance on top of them.

Ye Du, Yujun Shen, Haochen Wang et al.

Self-training has shown great potential in semi-supervised learning. Its core idea is to use the model learned on labeled data to generate pseudo-labels for unlabeled samples, and in turn teach itself. To obtain valid supervision, active attempts typically employ a momentum teacher for pseudo-label prediction yet observe the confirmation bias issue, where the incorrect predictions may provide wrong supervision signals and get accumulated in the training process. The primary cause of such a drawback is that the prevailing self-training framework acts as guiding the current state with previous knowledge, because the teacher is updated with the past student only. To alleviate this problem, we propose a novel self-training strategy, which allows the model to learn from the future. Concretely, at each training step, we first virtually optimize the student (i.e., caching the gradients without applying them to the model weights), then update the teacher with the virtual future student, and finally ask the teacher to produce pseudo-labels for the current student as the guidance. In this way, we manage to improve the quality of pseudo-labels and thus boost the performance. We also develop two variants of our future-self-training (FST) framework through peeping at the future both deeply (FST-D) and widely (FST-W). Taking the tasks of unsupervised domain adaptive semantic segmentation and semi-supervised semantic segmentation as the instances, we experimentally demonstrate the effectiveness and superiority of our approach under a wide range of settings. Code will be made publicly available.

Zhanchao Huang, Wei Li, Xiang-Gen Xia et al.

Arbitrary-oriented object detection (AOOD) has been widely applied to locate and classify objects with diverse orientations in remote sensing images. However, the inconsistent features for the localization and classification tasks in AOOD models may lead to ambiguity and low-quality object predictions, which constrains the detection performance. In this article, an AOOD method called task-wise sampling convolutions (TS-Conv) is proposed. TS-Conv adaptively samples task-wise features from respective sensitive regions and maps these features together in alignment to guide a dynamic label assignment for better predictions. Specifically, sampling positions of the localization convolution in TS-Conv are supervised by the oriented bounding box (OBB) prediction associated with spatial coordinates, while sampling positions and convolutional kernel of the classification convolution are designed to be adaptively adjusted according to different orientations for improving the orientation robustness of features. Furthermore, a dynamic task-consistent-aware label assignment (DTLA) strategy is developed to select optimal candidate positions and assign labels dynamically according to ranked task-aware scores obtained from TS-Conv. Extensive experiments on several public datasets covering multiple scenes, multimodal images, and multiple categories of objects demonstrate the effectiveness, scalability, and superior performance of the proposed TS-Conv.

Wei Li, Xing Wang, Xin Xia et al.

Vision Transformers have witnessed prevailing success in a series of vision tasks. However, these Transformers often rely on extensive computational costs to achieve high performance, which is burdensome to deploy on resource-constrained devices. To alleviate this issue, we draw lessons from depthwise separable convolution and imitate its ideology to design an efficient Transformer backbone, i.e., Separable Vision Transformer, abbreviated as SepViT. SepViT helps to carry out the local-global information interaction within and among the windows in sequential order via a depthwise separable self-attention. The novel window token embedding and grouped self-attention are employed to compute the attention relationship among windows with negligible cost and establish long-range visual interactions across multiple windows, respectively. Extensive experiments on general-purpose vision benchmarks demonstrate that SepViT can achieve a state-of-the-art trade-off between performance and latency. Among them, SepViT achieves 84.2% top-1 accuracy on ImageNet-1K classification while decreasing the latency by 40%, compared to the ones with similar accuracy (e.g., CSWin). Furthermore, SepViT achieves 51.0% mIoU on ADE20K semantic segmentation task, 47.9 AP on the RetinaNet-based COCO detection task, 49.4 box AP and 44.6 mask AP on Mask R-CNN-based COCO object detection and instance segmentation tasks.

Wei Li, Shuai Xiao, Tianhong Dai et al.

Ghosting artifacts, motion blur, and low fidelity in highlight are the main challenges in High Dynamic Range (HDR) imaging from multiple Low Dynamic Range (LDR) images. These issues come from using the medium-exposed image as the reference frame in previous methods. To deal with them, we propose to use the under-exposed image as the reference to avoid these issues. However, the heavy noise in dark regions of the under-exposed image becomes a new problem. Therefore, we propose a joint HDR and denoising pipeline, containing two sub-networks: (i) a pre-denoising network (PreDNNet) to adaptively denoise input LDRs by exploiting exposure priors; (ii) a pyramid cascading fusion network (PCFNet), introducing an attention mechanism and cascading structure in a multi-scale manner. To further leverage these two paradigms, we propose a selective and joint HDR and denoising (SJ-HD$^2$R) imaging framework, utilizing scenario-specific priors to conduct the path selection with an accuracy of more than 93.3$\%$. We create the first joint HDR and denoising benchmark dataset, which contains a variety of challenging HDR and denoising scenes and supports the switching of the reference image. Extensive experiment results show that our method achieves superior performance to previous methods.

Hao Wang, Zhanchao Huang, Zhengchao Chen et al.

Arbitrary-oriented object detection (AOOD) plays a significant role for image understanding in remote sensing scenarios. The existing AOOD methods face the challenges of ambiguity and high costs in angle representation. To this end, a multi-grained angle representation (MGAR) method, consisting of coarse-grained angle classification (CAC) and fine-grained angle regression (FAR), is proposed. Specifically, the designed CAC avoids the ambiguity of angle prediction by discrete angular encoding (DAE) and reduces complexity by coarsening the granularity of DAE. Based on CAC, FAR is developed to refine the angle prediction with much lower costs than narrowing the granularity of DAE. Furthermore, an Intersection over Union (IoU) aware FAR-Loss (IFL) is designed to improve accuracy of angle prediction using an adaptive re-weighting mechanism guided by IoU. Extensive experiments are performed on several public remote sensing datasets, which demonstrate the effectiveness of the proposed MGAR. Moreover, experiments on embedded devices demonstrate that the proposed MGAR is also friendly for lightweight deployments.

Xiaoyi Dong, Pan Zhang, Yuhang Zang et al. · pku

We introduce InternLM-XComposer2, a cutting-edge vision-language model excelling in free-form text-image composition and comprehension. This model goes beyond conventional vision-language understanding, adeptly crafting interleaved text-image content from diverse inputs like outlines, detailed textual specifications, and reference images, enabling highly customizable content creation. InternLM-XComposer2 proposes a Partial LoRA (PLoRA) approach that applies additional LoRA parameters exclusively to image tokens to preserve the integrity of pre-trained language knowledge, striking a balance between precise vision understanding and text composition with literary talent. Experimental results demonstrate the superiority of InternLM-XComposer2 based on InternLM2-7B in producing high-quality long-text multi-modal content and its exceptional vision-language understanding performance across various benchmarks, where it not only significantly outperforms existing multimodal models but also matches or even surpasses GPT-4V and Gemini Pro in certain assessments. This highlights its remarkable proficiency in the realm of multimodal understanding. The InternLM-XComposer2 model series with 7B parameters are publicly available at https://github.com/InternLM/InternLM-XComposer.

Junbo Qiao, Shaohui Lin, Yunlun Zhang et al.

Real-world image super-resolution (RISR) has received increased focus for improving the quality of SR images under unknown complex degradation. Existing methods rely on the heavy SR models to enhance low-resolution (LR) images of different degradation levels, which significantly restricts their practical deployments on resource-limited devices. In this paper, we propose a novel Dynamic Channel Splitting scheme for efficient Real-world Image Super-Resolution, termed DCS-RISR. Specifically, we first introduce the light degradation prediction network to regress the degradation vector to simulate the real-world degradations, upon which the channel splitting vector is generated as the input for an efficient SR model. Then, a learnable octave convolution block is proposed to adaptively decide the channel splitting scale for low- and high-frequency features at each block, reducing computation overhead and memory cost by offering the large scale to low-frequency features and the small scale to the high ones. To further improve the RISR performance, Non-local regularization is employed to supplement the knowledge of patches from LR and HR subspace with free-computation inference. Extensive experiments demonstrate the effectiveness of DCS-RISR on different benchmark datasets. Our DCS-RISR not only achieves the best trade-off between computation/parameter and PSNR/SSIM metric, and also effectively handles real-world images with different degradation levels.

Zihao Huang, Shoukang Hu, Guangcong Wang et al.

Existing research on avatar creation is typically limited to laboratory datasets, which require high costs against scalability and exhibit insufficient representation of the real world. On the other hand, the web abounds with off-the-shelf real-world human videos, but these videos vary in quality and require accurate annotations for avatar creation. To this end, we propose an automatic annotating pipeline with filtering protocols to curate these humans from the web. Our pipeline surpasses state-of-the-art methods on the EMDB benchmark, and the filtering protocols boost verification metrics on web videos. We then curate WildAvatar, a web-scale in-the-wild human avatar creation dataset extracted from YouTube, with $10000+$ different human subjects and scenes. WildAvatar is at least $10\times$ richer than previous datasets for 3D human avatar creation and closer to the real world. To explore its potential, we demonstrate the quality and generalizability of avatar creation methods on WildAvatar. We will publicly release our code, data source links and annotations to push forward 3D human avatar creation and other related fields for real-world applications.

Shimin Chen, Chen Chen, Wei Li et al.

Temporal action detection (TAD) aims to detect the semantic labels and boundaries of action instances in untrimmed videos. Current mainstream approaches are multi-step solutions, which fall short in efficiency and flexibility. In this paper, we propose a unified network for TAD, termed Faster-TAD, by re-purposing a Faster-RCNN like architecture. To tackle the unique difficulty in TAD, we make important improvements over the original framework. We propose a new Context-Adaptive Proposal Module and an innovative Fake-Proposal Generation Block. What's more, we use atomic action features to improve the performance. Faster-TAD simplifies the pipeline of TAD and gets remarkable performance on lots of benchmarks, i.e., ActivityNet-1.3 (40.01% mAP), HACS Segments (38.39% mAP), SoccerNet-Action Spotting (54.09% mAP). It outperforms existing single-network detector by a large margin.

Junbo Yin, Jianbing Shen, Runnan Chen et al.

Bird's eye view (BEV) representation has emerged as a dominant solution for describing 3D space in autonomous driving scenarios. However, objects in the BEV representation typically exhibit small sizes, and the associated point cloud context is inherently sparse, which leads to great challenges for reliable 3D perception. In this paper, we propose IS-Fusion, an innovative multimodal fusion framework that jointly captures the Instance- and Scene-level contextual information. IS-Fusion essentially differs from existing approaches that only focus on the BEV scene-level fusion by explicitly incorporating instance-level multimodal information, thus facilitating the instance-centric tasks like 3D object detection. It comprises a Hierarchical Scene Fusion (HSF) module and an Instance-Guided Fusion (IGF) module. HSF applies Point-to-Grid and Grid-to-Region transformers to capture the multimodal scene context at different granularities. IGF mines instance candidates, explores their relationships, and aggregates the local multimodal context for each instance. These instances then serve as guidance to enhance the scene feature and yield an instance-aware BEV representation. On the challenging nuScenes benchmark, IS-Fusion outperforms all the published multimodal works to date. Code is available at: https://github.com/yinjunbo/IS-Fusion.

Yuxiang Zhang, Shunlin Liang, Wenyuan Li et al.

One of the primary objectives of satellite remote sensing is to capture the complex dynamics of the Earth environment, which encompasses tasks such as reconstructing continuous cloud-free time series images, detecting land cover changes, and forecasting future surface evolution. However, existing methods typically require specialized models tailored to different tasks, lacking unified modeling of spatiotemporal features across multiple time series tasks. In this paper, we propose a Unified Time Series Generative Model (UniTS), a general framework applicable to various time series tasks, including time series reconstruction, time series cloud removal, time series semantic change detection, and time series forecasting. Based on the flow matching generative paradigm, UniTS constructs a deterministic evolution path from noise to targets under the guidance of task-specific conditions, achieving unified modeling of spatiotemporal representations for multiple tasks. The UniTS architecture consists of a diffusion transformer with spatio-temporal blocks, where we design an Adaptive Condition Injector (ACor) to enhance the model's conditional perception of multimodal inputs, enabling high-quality controllable generation. Additionally, we design a Spatiotemporal-aware Modulator (STM) to improve the ability of spatio-temporal blocks to capture complex spatiotemporal dependencies. Furthermore, we construct two high-quality multimodal time series datasets, TS-S12 and TS-S12CR, filling the gap of benchmark datasets for time series cloud removal and forecasting tasks. Extensive experiments demonstrate that UniTS exhibits exceptional generative and cognitive capabilities in both low-level and high-level time series tasks. It significantly outperforms existing methods, particularly when facing challenges such as severe cloud contamination, modality absence, and forecasting phenological variations.

Shimin Chen, Wei Li, Chen Chen et al.

In this paper, we introduce a novel large-scale video dataset dubbed MM-SEAL for multi-person multi-grained spatio-temporal action localization among human daily life. We are the first to propose a new benchmark for multi-person spatio-temporal complex activity localization, where complex semantic and long duration bring new challenges to localization tasks. We observe that limited atomic actions can be combined into many complex activities. MM-SEAL provides both atomic action and complex activity annotations, producing 111.7k atomic actions spanning 172 action categories and 17.7k complex activities spanning 200 activity categories. We explore the relationship between atomic actions and complex activities, finding that atomic action features can improve the complex activity localization performance. Also, we propose a new network which generates temporal proposals and labels simultaneously, termed Faster-TAD. Finally, our evaluations show that visual features pretrained on MM-SEAL can improve the performance on other action localization benchmarks. We will release the dataset and the project code upon publication of the paper.

Yinjian Wang, Wei Li, Yuanyuan Gui et al.

Hyperspectral super-resolution is commonly accomplished by the fusing of a hyperspectral imaging of low spatial resolution with a multispectral image of high spatial resolution, and many tensor-based approaches to this task have been recently proposed. Yet, it is assumed in such tensor-based methods that the spatial-blurring operation that creates the observed hyperspectral image from the desired super-resolved image is separable into independent horizontal and vertical blurring. Recent work has argued that such separable spatial degradation is ill-equipped to model the operation of real sensors which may exhibit, for example, anisotropic blurring. To accommodate this fact, a generalized tensor formulation based on a Kronecker decomposition is proposed to handle any general spatial-degradation matrix, including those that are not separable as previously assumed. Analysis of the generalized formulation reveals conditions under which exact recovery of the desired super-resolved image is guaranteed, and a practical algorithm for such recovery, driven by a blockwise-group-sparsity regularization, is proposed. Extensive experimental results demonstrate that the proposed generalized tensor approach outperforms not only traditional matrix-based techniques but also state-of-the-art tensor-based methods; the gains with respect to the latter are especially significant in cases of anisotropic spatial blurring.

Hanzheng Wang, Wei Li, Xiang-Gen Xia et al.

Hyperspectral object tracking (HOT) has exhibited potential in various applications, particularly in scenes where objects are camouflaged. Existing trackers can effectively retrieve objects via band regrouping because of the bias in existing HOT datasets, where most objects tend to have distinguishing visual appearances rather than spectral characteristics. This bias allows the tracker to directly use the visual features obtained from the false-color images generated by hyperspectral images without the need to extract spectral features. To tackle this bias, we find that the tracker should focus on the spectral information when object appearance is unreliable. Thus, we provide a new task called hyperspectral camouflaged object tracking (HCOT) and meticulously construct a large-scale HCOT dataset, termed BihoT, which consists of 41,912 hyperspectral images covering 49 video sequences. The dataset covers various artificial camouflage scenes where objects have similar appearances, diverse spectrums, and frequent occlusion, making it a very challenging dataset for HCOT. Besides, a simple but effective baseline model, named spectral prompt-based distractor-aware network (SPDAN), is proposed, comprising a spectral embedding network (SEN), a spectral prompt-based backbone network (SPBN), and a distractor-aware module (DAM). Specifically, the SEN extracts spectral-spatial features via 3-D and 2-D convolutions. Then, the SPBN fine-tunes powerful RGB trackers with spectral prompts and alleviates the insufficiency of training samples. Moreover, the DAM utilizes a novel statistic to capture the distractor caused by occlusion from objects and background. Extensive experiments demonstrate that our proposed SPDAN achieves state-of-the-art performance on the proposed BihoT and other HOT datasets.

Yuxiang Zhang, Wei Li, Wen Jia et al.

Utilizing hyperspectral remote sensing technology enables the extraction of fine-grained land cover classes. Typically, satellite or airborne images used for training and testing are acquired from different regions or times, where the same class has significant spectral shifts in different scenes. In this paper, we propose a Bi-directional Domain Adaptation (BiDA) framework for cross-domain hyperspectral image (HSI) classification, which focuses on extracting both domain-invariant features and domain-specific information in the independent adaptive space, thereby enhancing the adaptability and separability to the target scene. In the proposed BiDA, a triple-branch transformer architecture (the source branch, target branch, and coupled branch) with semantic tokenizer is designed as the backbone. Specifically, the source branch and target branch independently learn the adaptive space of source and target domains, a Coupled Multi-head Cross-attention (CMCA) mechanism is developed in coupled branch for feature interaction and inter-domain correlation mining. Furthermore, a bi-directional distillation loss is designed to guide adaptive space learning using inter-domain correlation. Finally, we propose an Adaptive Reinforcement Strategy (ARS) to encourage the model to focus on specific generalized feature extraction within both source and target scenes in noise condition. Experimental results on cross-temporal/scene airborne and satellite datasets demonstrate that the proposed BiDA performs significantly better than some state-of-the-art domain adaptation approaches. In the cross-temporal tree species classification task, the proposed BiDA is more than 3\%$\sim$5\% higher than the most advanced method. The codes will be available from the website: https://github.com/YuxiangZhang-BIT/IEEE_TCSVT_BiDA.

Yudong Yang, Jimin Zhuang, Guangzhi Sun et al.

Audio often serves as an auxiliary modality in video understanding tasks of audio-visual large language models (LLMs), merely assisting in the comprehension of visual information. However, a thorough understanding of videos significantly depends on auditory information, as audio offers critical context, emotional cues, and semantic meaning that visual data alone often lacks. This paper proposes an audio-centric video understanding benchmark (AVUT) to evaluate the video comprehension capabilities of multimodal LLMs with a particular focus on auditory information. AVUT introduces a suite of carefully designed audio-centric tasks, holistically testing the understanding of both audio content and audio-visual interactions in videos. Moreover, this work points out the text shortcut problem that largely exists in other benchmarks where the correct answer can be found from question text alone without needing videos. AVUT addresses this problem by proposing a answer permutation-based filtering mechanism. A thorough evaluation across a diverse range of open-source and proprietary multimodal LLMs is performed, followed by the analyses of deficiencies in audio-visual LLMs. Demos and data are available at https://github.com/lark-png/AVUT.

Shaohang Wei, Wei Li, Feifan Song et al. · pku

Temporal reasoning is pivotal for Large Language Models (LLMs) to comprehend the real world. However, existing works neglect the real-world challenges for temporal reasoning: (1) intensive temporal information, (2) fast-changing event dynamics, and (3) complex temporal dependencies in social interactions. To bridge this gap, we propose a multi-level benchmark TIME, designed for temporal reasoning in real-world scenarios. TIME consists of 38,522 QA pairs, covering 3 levels with 11 fine-grained sub-tasks. This benchmark encompasses 3 sub-datasets reflecting different real-world challenges: TIME-Wiki, TIME-News, and TIME-Dial. We conduct extensive experiments on reasoning models and non-reasoning models. And we conducted an in-depth analysis of temporal reasoning performance across diverse real-world scenarios and tasks, and summarized the impact of test-time scaling on temporal reasoning capabilities. Additionally, we release TIME-Lite, a human-annotated subset to foster future research and standardized evaluation in temporal reasoning. The code is available at https://github.com/sylvain-wei/TIME , the dataset is available at https://huggingface.co/datasets/SylvainWei/TIME , and the project page link is https://sylvain-wei.github.io/TIME/ .

Mingjia Li, Shuang Li, Tongrui Su et al.

Capitalizing on the complementary advantages of generative and discriminative models has always been a compelling vision in machine learning, backed by a growing body of research. This work discloses the hidden semantic structure within score-based generative models, unveiling their potential as effective discriminative priors. Inspired by our theoretical findings, we propose DUSA to exploit the structured semantic priors underlying diffusion score to facilitate the test-time adaptation of image classifiers or dense predictors. Notably, DUSA extracts knowledge from a single timestep of denoising diffusion, lifting the curse of Monte Carlo-based likelihood estimation over timesteps. We demonstrate the efficacy of our DUSA in adapting a wide variety of competitive pre-trained discriminative models on diverse test-time scenarios. Additionally, a thorough ablation study is conducted to dissect the pivotal elements in DUSA. Code is publicly available at https://github.com/BIT-DA/DUSA.

Yongxu Wang, Weiyun Yi, Xinhao Kong et al.

With the rapid development of embodied intelligence, leveraging large-scale human data for high-level imitation learning on humanoid robots has become a focal point of interest in both academia and industry. However, applying humanoid robots to precision operation domains remains challenging due to the complexities they face in perception and control processes, the long-standing physical differences in morphology and actuation mechanisms between humanoid robots and humans, and the lack of task-relevant features obtained from egocentric vision. To address the issue of covariate shift in imitation learning, this paper proposes an imitation learning algorithm tailored for humanoid robots. By focusing on the primary task objectives, filtering out background information, and incorporating channel feature fusion with spatial attention mechanisms, the proposed algorithm suppresses environmental disturbances and utilizes a dynamic weight update strategy to significantly improve the success rate of humanoid robots in accomplishing target tasks. Experimental results demonstrate that the proposed method exhibits robustness and scalability across various typical task scenarios, providing new ideas and approaches for autonomous learning and control in humanoid robots. The project will be open-sourced on GitHub.

Zhibin Wang, Yanxin Cai, Jiayi Zhou et al.

The growing field of remote sensing faces a challenge: the ever-increasing size and volume of imagery data are exceeding the storage and transmission capabilities of satellite platforms. Efficient compression of remote sensing imagery is a critical solution to alleviate these burdens on satellites. However, existing compression methods are often too computationally expensive for satellites. With the continued advancement of compressed sensing theory, single-pixel imaging emerges as a powerful tool that brings new possibilities for on-orbit image compression. However, it still suffers from prolonged imaging times and the inability to perform high-resolution imaging, hindering its practical application. This paper advances the study of compressed sensing in remote sensing image compression, proposing Block Modulated Imaging (BMI). By requiring only a single exposure, BMI significantly enhances imaging acquisition speeds. Additionally, BMI obviates the need for digital micromirror devices and surpasses limitations in image resolution. Furthermore, we propose a novel decoding network specifically designed to reconstruct images compressed under the BMI framework. Leveraging the gated 3D convolutions and promoting efficient information flow across stages through a Two-Way Cross-Attention module, our decoding network exhibits demonstrably superior reconstruction performance. Extensive experiments conducted on multiple renowned remote sensing datasets unequivocally demonstrate the efficacy of our proposed method. To further validate its practical applicability, we developed and tested a prototype of the BMI-based camera, which has shown promising potential for on-orbit image compression. The code is available at https://github.com/Johnathan218/BMNet.

Yuhang Zang, Wei Li, Jun Han et al.

Recent Multimodal Large Language Models (MLLMs) are remarkable in vision-language tasks, such as image captioning and question answering, but lack the essential perception ability, i.e., object detection. In this work, we address this limitation by introducing a novel research problem of contextual object detection -- understanding visible objects within different human-AI interactive contexts. Three representative scenarios are investigated, including the language cloze test, visual captioning, and question answering. Moreover, we present ContextDET, a unified multimodal model that is capable of end-to-end differentiable modeling of visual-language contexts, so as to locate, identify, and associate visual objects with language inputs for human-AI interaction. Our ContextDET involves three key submodels: (i) a visual encoder for extracting visual representations, (ii) a pre-trained LLM for multimodal context decoding, and (iii) a visual decoder for predicting bounding boxes given contextual object words. The new generate-then-detect framework enables us to detect object words within human vocabulary. Extensive experiments show the advantages of ContextDET on our proposed CODE benchmark, open-vocabulary detection, and referring image segmentation. Github: https://github.com/yuhangzang/ContextDET.

Kehe WU, Zuge Chen, Qi MA et al.

As the scale of object detection dataset is smaller than that of image recognition dataset ImageNet, transfer learning has become a basic training method for deep learning object detection models, which will pretrain the backbone network of object detection model on ImageNet dataset to extract features for classification and localization subtasks. However, the classification task focuses on the salient region features of object, while the location task focuses on the edge features of object, so there is certain deviation between the features extracted by pretrained backbone network and the features used for localization task. In order to solve this problem, a decoupled self attention(DSA) module is proposed for one stage object detection models in this paper. DSA includes two decoupled self-attention branches, so it can extract appropriate features for different tasks. It is located between FPN and head networks of subtasks, so it is used to extract global features based on FPN fused features for different tasks independently. Although the network of DSA module is simple, but it can effectively improve the performance of object detection, also it can be easily embedded in many detection models. Our experiments are based on the representative one-stage detection model RetinaNet. In COCO dataset, when ResNet50 and ResNet101 are used as backbone networks, the detection performances can be increased by 0.4% AP and 0.5% AP respectively. When DSA module and object confidence task are applied in RetinaNet together, the detection performances based on ResNet50 and ResNet101 can be increased by 1.0% AP and 1.4% AP respectively. The experiment results show the effectiveness of DSA module. Code is at: https://github.com/chenzuge1/DSANet.git.

Wu Kehe, Chen Zuge, Zhang Xiaoliang et al.

RetinaNet proposed Focal Loss for classification task and improved one-stage detectors greatly. However, there is still a gap between it and two-stage detectors. We analyze the prediction of RetinaNet and find that the misalignment of classification and localization is the main factor. Most of predicted boxes, whose IoU with ground-truth boxes are greater than 0.5, while their classification scores are lower than 0.5, which shows that the classification task still needs to be optimized. In this paper we proposed an object confidence task for this problem, and it shares features with classification task. This task uses IoUs between samples and ground-truth boxes as targets, and it only uses losses of positive samples in training, which can increase loss weight of positive samples in classification task training. Also the joint of classification score and object confidence will be used to guide NMS. Our method can not only improve classification task, but also ease misalignment of classification and localization. To evaluate the effectiveness of this method, we show our experiments on MS COCO 2017 dataset. Without whistles and bells, our method can improve AP by 0.7% and 1.0% on COCO validation dataset with ResNet50 and ResNet101 respectively at same training configs, and it can achieve 38.4% AP with two times training time. Code is at: http://github.com/chenzuge1/RetinaNet-Conf.git.

Bo Yang Hsueh, Wei Li, I-Chen Wu

Learning rate scheduler has been a critical issue in the deep neural network training. Several schedulers and methods have been proposed, including step decay scheduler, adaptive method, cosine scheduler and cyclical scheduler. This paper proposes a new scheduling method, named hyperbolic-tangent decay (HTD). We run experiments on several benchmarks such as: ResNet, Wide ResNet and DenseNet for CIFAR-10 and CIFAR-100 datasets, LSTM for PAMAP2 dataset, ResNet on ImageNet and Fashion-MNIST datasets. In our experiments, HTD outperforms step decay and cosine scheduler in nearly all cases, while requiring less hyperparameters than step decay, and more flexible than cosine scheduler. Code is available at https://github.com/BIGBALLON/HTD.

Quan Zhang, Xiaoyu Liu, Wei Li et al.

In image restoration (IR), leveraging semantic priors from segmentation models has been a common approach to improve performance. The recent segment anything model (SAM) has emerged as a powerful tool for extracting advanced semantic priors to enhance IR tasks. However, the computational cost of SAM is prohibitive for IR, compared to existing smaller IR models. The incorporation of SAM for extracting semantic priors considerably hampers the model inference efficiency. To address this issue, we propose a general framework to distill SAM's semantic knowledge to boost exiting IR models without interfering with their inference process. Specifically, our proposed framework consists of the semantic priors fusion (SPF) scheme and the semantic priors distillation (SPD) scheme. SPF fuses two kinds of information between the restored image predicted by the original IR model and the semantic mask predicted by SAM for the refined restored image. SPD leverages a self-distillation manner to distill the fused semantic priors to boost the performance of original IR models. Additionally, we design a semantic-guided relation (SGR) module for SPD, which ensures semantic feature representation space consistency to fully distill the priors. We demonstrate the effectiveness of our framework across multiple IR models and tasks, including deraining, deblurring, and denoising.

Xuhui Jiang, Yinghan Shen, Zhichao Shi et al.

Entity Alignment (EA) is vital for integrating diverse knowledge graph (KG) data, playing a crucial role in data-driven AI applications. Traditional EA methods primarily rely on comparing entity embeddings, but their effectiveness is constrained by the limited input KG data and the capabilities of the representation learning techniques. Against this backdrop, we introduce ChatEA, an innovative framework that incorporates large language models (LLMs) to improve EA. To address the constraints of limited input KG data, ChatEA introduces a KG-code translation module that translates KG structures into a format understandable by LLMs, thereby allowing LLMs to utilize their extensive background knowledge to improve EA accuracy. To overcome the over-reliance on entity embedding comparisons, ChatEA implements a two-stage EA strategy that capitalizes on LLMs' capability for multi-step reasoning in a dialogue format, thereby enhancing accuracy while preserving efficiency. Our experimental results verify ChatEA's superior performance, highlighting LLMs' potential in facilitating EA tasks.

Ziyu Liu, Zeyi Sun, Yuhang Zang et al.

CLIP (Contrastive Language-Image Pre-training) uses contrastive learning from noise image-text pairs to excel at recognizing a wide array of candidates, yet its focus on broad associations hinders the precision in distinguishing subtle differences among fine-grained items. Conversely, Multimodal Large Language Models (MLLMs) excel at classifying fine-grained categories, thanks to their substantial knowledge from pre-training on web-level corpora. However, the performance of MLLMs declines with an increase in category numbers, primarily due to growing complexity and constraints of limited context window size. To synergize the strengths of both approaches and enhance the few-shot/zero-shot recognition abilities for datasets characterized by extensive and fine-grained vocabularies, this paper introduces RAR, a Retrieving And Ranking augmented method for MLLMs. We initially establish a multi-modal retriever based on CLIP to create and store explicit memory for different categories beyond the immediate context window. During inference, RAR retrieves the top-k similar results from the memory and uses MLLMs to rank and make the final predictions. Our proposed approach not only addresses the inherent limitations in fine-grained recognition but also preserves the model's comprehensive knowledge base, significantly boosting accuracy across a range of vision-language recognition tasks. Notably, our approach demonstrates a significant improvement in performance on 5 fine-grained visual recognition benchmarks, 11 few-shot image recognition datasets, and the 2 object detection datasets under the zero-shot recognition setting.