Temporal action detection aims to locate and classify actions in untrimmed videos. While recent works focus on designing powerful feature processors for pre-trained representations, they often overlook the inherent noise and redundancy within these features. Large-scale pre-trained video encoders tend to introduce background clutter and irrelevant semantics, leading to context confusion and imprecise boundaries. To address this, we propose a frequency-aware decoupling network that improves action discriminability by filtering out noisy semantics captured by pre-trained models. Specifically, we introduce an adaptive temporal decoupling scheme that suppresses irrelevant information while preserving fine-grained atomic action details, yielding more task-specific representations. In addition, we enhance inter-frame modeling by capturing temporal variations to better distinguish actions from background redundancy. Furthermore, we present a long-short-term category-aware relation network that jointly models local transitions and long-range dependencies, improving localization precision. The refined atomic features and frequency-guided dynamics are fed into a standard detection head to produce accurate action predictions. Extensive experiments on THUMOS14, HACS, and ActivityNet-1.3 show that our method, powered by InternVideo2-6B features, achieves state-of-the-art performance on temporal action detection benchmarks.
Humans and other animals readily generalize abstract relations, such as recognizing constant in shape or color, whereas neural networks struggle. To investigate how neural networks generalize abstract relations, we introduce SimplifiedRPM, a novel benchmark for systematic evaluation. In parallel, we conduct human experiments to benchmark relational difficulty, enabling direct model-human comparisons. Testing four architectures--ResNet-50, Vision Transformer, Wild Relation Network, and Scattering Compositional Learner (SCL)--we find that SCL best aligns with human behavior and generalizes best. Building on a geometric theory of neural representations, we show representational geometries that predict generalization. Layer-wise analysis reveals distinct relational reasoning strategies across models and suggests a trade-off where unseen rule representations compress into training-shaped subspaces. Guided by our geometric perspective, we propose and evaluate SNRloss, a novel objective balancing representation geometry. Our findings offer geometric insights into how neural networks generalize abstract relations, paving the way for more human-like visual reasoning in AI.
This paper introduces a new problem, Causal Abductive Reasoning on Video Events (CARVE), which involves identifying causal relationships between events in a video and generating hypotheses about causal chains that account for the occurrence of a target event. To facilitate research in this direction, we create two new benchmark datasets with both synthetic and realistic videos, accompanied by trigger-target labels generated through a novel counterfactual synthesis approach. To explore the challenge of solving CARVE, we present a Causal Event Relation Network (CERN) that examines the relationships between video events in temporal and semantic spaces to efficiently determine the root-cause trigger events. Through extensive experiments, we demonstrate the critical roles of event relational representation learning and interaction modeling in solving video causal reasoning challenges. The introduction of the CARVE task, along with the accompanying datasets and the CERN framework, will advance future research on video causal reasoning and significantly facilitate various applications, including video surveillance, root-cause analysis and movie content management.
This paper presents the ARN-LSTM architecture, a novel multi-stream action recognition model designed to address the challenge of simultaneously capturing spatial motion and temporal dynamics in action sequences. Traditional methods often focus solely on spatial or temporal features, limiting their ability to comprehend complex human activities fully. Our proposed model integrates joint, motion, and temporal information through a multi-stream fusion architecture. Specifically, it comprises a jointstream for extracting skeleton features, a temporal stream for capturing dynamic temporal features, and an ARN-LSTM block that utilizes Time-Distributed Long Short-Term Memory (TD-LSTM) layers followed by an Attention Relation Network (ARN) to model temporal relations. The outputs from these streams are fused in a fully connected layer to provide the final action prediction. Evaluations on the NTU RGB+D 60 and NTU RGB+D 120 datasets outperform the superior performance of our model, particularly in group activity recognition.
LLM-based agents have been widely applied as personal assistants, capable of memorizing information from user messages and responding to personal queries. However, there still lacks an objective and automatic evaluation on their memory capability, largely due to the challenges in constructing reliable questions and answers (QAs) according to user messages. In this paper, we propose MemSim, a Bayesian simulator designed to automatically construct reliable QAs from generated user messages, simultaneously keeping their diversity and scalability. Specifically, we introduce the Bayesian Relation Network (BRNet) and a causal generation mechanism to mitigate the impact of LLM hallucinations on factual information, facilitating the automatic creation of an evaluation dataset. Based on MemSim, we generate a dataset in the daily-life scenario, named MemDaily, and conduct extensive experiments to assess the effectiveness of our approach. We also provide a benchmark for evaluating different memory mechanisms in LLM-based agents with the MemDaily dataset. To benefit the research community, we have released our project at https://github.com/nuster1128/MemSim.
Recent work demonstrated Transformers' ability to efficiently copy strings of exponential sizes, distinguishing them from other architectures. We present the Causal Relation Network (CausalRN), an all-MLP sequence modeling architecture that can match Transformers on the copying task. Extending Relation Networks (RNs), we implemented key innovations to support autoregressive sequence modeling while maintaining computational feasibility. We discovered that exponentially-activated RNs are reducible to linear time complexity, and pre-activation normalization induces an infinitely growing memory pool, similar to a KV cache. In ablation study, we found both exponential activation and pre-activation normalization are indispensable for Transformer-level copying. Our findings provide new insights into what actually constitutes strong in-context retrieval.
Federated learning enables distributed clients to collaborate on training while storing their data locally to protect client privacy. However, due to the heterogeneity of data, models, and devices, the final global model may need to perform better for tasks on each client. Communication bottlenecks, data heterogeneity, and model heterogeneity have been common challenges in federated learning. In this work, we considered a label distribution skew problem, a type of data heterogeneity easily overlooked. In the context of classification, we propose a personalized federated learning approach called pFedPM. In our process, we replace traditional gradient uploading with feature uploading, which helps reduce communication costs and allows for heterogeneous client models. These feature representations play a role in preserving privacy to some extent. We use a hyperparameter $a$ to mix local and global features, which enables us to control the degree of personalization. We also introduced a relation network as an additional decision layer, which provides a non-linear learnable classifier to predict labels. Experimental results show that, with an appropriate setting of $a$, our scheme outperforms several recent FL methods on MNIST, FEMNIST, and CRIFAR10 datasets and achieves fewer communications.
Inductive knowledge graph completion (KGC) aims to infer the missing relation for a set of newly-coming entities that never appeared in the training set. Such a setting is more in line with reality, as real-world KGs are constantly evolving and introducing new knowledge. Recent studies have shown promising results using message passing over subgraphs to embed newly-coming entities for inductive KGC. However, the inductive capability of these methods is usually limited by two key issues. (i) KGC always suffers from data sparsity, and the situation is even exacerbated in inductive KGC where new entities often have few or no connections to the original KG. (ii) Cold-start problem. It is over coarse-grained for accurate KG reasoning to generate representations for new entities by gathering the local information from few neighbors. To this end, we propose a novel iNfOmax RelAtion Network, namely NORAN, for inductive KG completion. It aims to mine latent relation patterns for inductive KG completion. Specifically, by centering on relations, NORAN provides a hyper view towards KG modeling, where the correlations between relations can be naturally captured as entity-independent logical evidence to conduct inductive KGC. Extensive experiment results on five benchmarks show that our framework substantially outperforms the state-of-the-art KGC methods.
Multi-person pose estimation (MPPE) presents a formidable yet crucial challenge in computer vision. Most existing methods predominantly concentrate on isolated interaction either between instances or joints, which is inadequate for scenarios demanding concurrent localization of both instances and joints. This paper introduces a novel CNN-based single-stage method, named Dual-path Hierarchical Relation Network (DHRNet), to extract instance-to-joint and joint-to-instance interactions concurrently. Specifically, we design a dual-path interaction modeling module (DIM) that strategically organizes cross-instance and cross-joint interaction modeling modules in two complementary orders, enriching interaction information by integrating merits from different correlation modeling branches. Notably, DHRNet excels in joint localization by leveraging information from other instances and joints. Extensive evaluations on challenging datasets, including COCO, CrowdPose, and OCHuman datasets, showcase DHRNet's state-of-the-art performance. The code will be released at https://github.com/YHDang/dhrnet-multi-pose-estimation.
Co-occurrent visual patterns suggest that pixel relation modeling facilitates dense prediction tasks, which inspires the development of numerous context modeling paradigms, \emph{e.g.}, multi-scale-driven and similarity-driven context schemes. Despite the impressive results, these existing paradigms often suffer from inadequate or ineffective contextual information aggregation due to reliance on large amounts of predetermined priors. To alleviate the issues, we propose a novel \textbf{I}ntervention-\textbf{D}riven \textbf{R}elation \textbf{Net}work (\textbf{IDRNet}), which leverages a deletion diagnostics procedure to guide the modeling of contextual relations among different pixels. Specifically, we first group pixel-level representations into semantic-level representations with the guidance of pseudo labels and further improve the distinguishability of the grouped representations with a feature enhancement module. Next, a deletion diagnostics procedure is conducted to model relations of these semantic-level representations via perceiving the network outputs and the extracted relations are utilized to guide the semantic-level representations to interact with each other. Finally, the interacted representations are utilized to augment original pixel-level representations for final predictions. Extensive experiments are conducted to validate the effectiveness of IDRNet quantitatively and qualitatively. Notably, our intervention-driven context scheme brings consistent performance improvements to state-of-the-art segmentation frameworks and achieves competitive results on popular benchmark datasets, including ADE20K, COCO-Stuff, PASCAL-Context, LIP, and Cityscapes. Code is available at \url{https://github.com/SegmentationBLWX/sssegmentation}.
How to select relevant key objects and reason about the complex relationships cross vision and linguistic domain are two key issues in many multi-modality applications such as visual question answering (VQA). In this work, we incorporate the visual commonsense information and propose a heterogeneous graph contrastive learning method to better finish the visual reasoning task. Our method is designed as a plug-and-play way, so that it can be quickly and easily combined with a wide range of representative methods. Specifically, our model contains two key components: the Commonsense-based Contrastive Learning and the Graph Relation Network. Using contrastive learning, we guide the model concentrate more on discriminative objects and relevant visual commonsense attributes. Besides, thanks to the introduction of the Graph Relation Network, the model reasons about the correlations between homogeneous edges and the similarities between heterogeneous edges, which makes information transmission more effective. Extensive experiments on four benchmarks show that our method greatly improves seven representative VQA models, demonstrating its effectiveness and generalizability.
In the process of building (structural learning) a probabilistic graphical model from a set of observed data, the directional, cyclic dependencies between the random variables of the model are often found. Existing graphical models such as Bayesian and Markov networks can reflect such dependencies. However, this requires complicating those models, such as adding additional variables or dividing the model graph into separate subgraphs. Herein, we describe a probabilistic graphical model - probabilistic relation network - that allows the direct capture of directional cyclic dependencies during structural learning. This model is based on the simple idea that each sample of the observed data can be represented by an arbitrary graph (structured outcome), which reflects the structure of the dependencies of the variables included in the sample. Each of the outcomes contains only a part of the graphical model structure; however, a complete graph of the probabilistic model is obtained by combining different outcomes. Such a graph, unlike Bayesian and Markov networks, can be directed and can have cycles. We explored the full joint distribution and conditional distribution and conditional independence properties of variables in the proposed model. We defined the algorithms for constructing of the model from the dataset and for calculating the conditional and full joint distributions. We also performed a numerical comparison with Bayesian and Markov networks. This model does not violate the probability axioms, and it supports learning from observed data. Notably, it supports probabilistic inference, making it a prospective tool in data analysis and in expert and design-making applications.
Few-shot learning or meta-learning leverages the data scarcity problem in machine learning. Traditionally, training data requires a multitude of samples and labeling for supervised learning. To address this issue, we propose a one-shot unsupervised meta-learning to learn the latent representation of the training samples. We use augmented samples as the query set during the training phase of the unsupervised meta-learning. A temperature-scaled cross-entropy loss is used in the inner loop of meta-learning to prevent overfitting during unsupervised learning. The learned parameters from this step are applied to the targeted supervised meta-learning in a transfer-learning fashion for initialization and fast adaptation with improved accuracy. The proposed method is model agnostic and can aid any meta-learning model to improve accuracy. We use model agnostic meta-learning (MAML) and relation network (RN) on Omniglot and mini-Imagenet datasets to demonstrate the performance of the proposed method. Furthermore, a meta-learning model with the proposed initialization can achieve satisfactory accuracy with significantly fewer training samples.
Geospatial object segmentation, a fundamental Earth vision task, always suffers from scale variation, the larger intraclass variance of background, and foreground-background imbalance in high spatial resolution (HSR) remote sensing imagery. Generic semantic segmentation methods mainly focus on the scale variation in natural scenarios. However, the other two problems are insufficiently considered in large area Earth observation scenarios. In this paper, we propose a foreground-aware relation network (FarSeg++) from the perspectives of relation-based, optimizationbased, and objectness-based foreground modeling, alleviating the above two problems. From the perspective of the relations, the foreground-scene relation module improves the discrimination of the foreground features via the foreground-correlated contexts associated with the object-scene relation. From the perspective of optimization, foreground-aware optimization is proposed to focus on foreground examples and hard examples of the background during training to achieve a balanced optimization. Besides, from the perspective of objectness, a foreground-aware decoder is proposed to improve the objectness representation, alleviating the objectness prediction problem that is the main bottleneck revealed by an empirical upper bound analysis. We also introduce a new large-scale high-resolution urban vehicle segmentation dataset to verify the effectiveness of the proposed method and push the development of objectness prediction further forward. The experimental results suggest that FarSeg++ is superior to the state-of-the-art generic semantic segmentation methods and can achieve a better trade-off between speed and accuracy.
Recently, few-shot learning (FSL) has exhibited great potential in the hyperspectral image (HSI) classification due to its promising performance under a few training samples. Although existing FSL methods have achieved great success, some limitations can still be witnessed. On the one hand, current methods mainly rely on a single metric to identify, which cannot effectively represent the class distribution with few labeled samples. On the other hand, existing methods usually only use the last deep feature of the feature extractor, which may lead to the under-utilization of scarce labeled samples. To overcome the above issues, a novel multistage relation network with dual-metric (DM-MRN) is proposed for few-shot HSI classification. First, a sample recombination strategy is designed to increase the variety of classification tasks in the training period. Second, an embedding module is employed to extract deep features of the input image patches. Third, we propose two relation modules: image-to-class (I2C) block and image-to-image (I2I) block. The I2C block is designed to compute the I2C-level relation score between second-order features, and the I2I block is conceived to generate the I2I-level relation score between first-order features. Finally, DM-MRN is constructed by integrating one embedding module, two I2C blocks, and one I2I block. In addition, an adaptive weighting strategy is designed to fuse the obtained relation scores, and classification can be achieved by assigning each query sample to the class with the highest value of the fused relation score. Extensive experiments carried out on five popular HSI datasets demonstrate that the proposed method outperforms other traditional and advanced models under few training samples in terms of classification accuracy and generalization performance, i.e., the performance improvement in terms of OA is around 0.30%–27.98% under ten labeled samples per class.
Previous personalized hashtag recommendations have been able to recommend suitable hashtags for a given microblog. Despite their performance improvement, we argue that three challenges remain unexplored. First, prior studies capture user interests solely from user-hashtag interactions that are directly connected (i.e., first-order relations), making them unable to deal with multiple user behaviors, including user-user social and hashtag-hashtag co-occurrence, and also restrict relations from similar users that are indirectly connected (i.e., high-order relations). Second, previous works personalize content at the microblog level, ignoring the personalized aspects that users have for each word in the microblog. Third, past studies capture correlations among hashtags in the same microblog from only the left-side correlations, restricting the right-side correlations. In this paper, we propose a novel integral model for personalized hashtag recommendation named PAC-MAN, which explores high-order multiple relations to model fruitful user and hashtag representation before fusing with word representation for word-level personalization and integrating with sequenceless hashtag correlation for the recommendation. First, to derive fruitful user and hashtag representation from higher-order multiple relations, Multi-relational Attentive Network (MAN) applies GNN to jointly capture relations on three communities: (1) user-hashtag interaction; (2) user-user social; and (3) hashtag-hashtag co-occurrence. Second, to personalize content at a word level, Person-And-Content based BERT (PAC) extends BERT to input not only word representations from the microblog but also the fruitful user representation from MAN, allowing each word to be fused with user aspects. Finally, to capture sequenceless hashtag correlations, the fruitful hashtag representations from MAN that contain the hashtag’s community perspectives are inserted into BERT to integrate with the hashtag’s word-semantic perspectives, and a hashtag prediction task is then conducted under the mask concept, enabling hashtag correlations to be obtained from both left and right sides without sequence constraints. Extensive experiments on the Twitter dataset demonstrate that PAC-MAN consistently outperforms state-of-the-art methods, including neural network based and traditional graph based methods, over precision, recall, and F1-score metrics.
Recent state-of-the-art methods for HOI detection typically build on transformer architectures with two decoder branches, one for human-object pair detection and the other for interaction classification. Such disentangled transformers, however, may suffer from insufficient context exchange between the branches and lead to a lack of context information for relational reasoning, which is critical in discovering HOI instances. In this work, we propose the multiplex relation network (MUREN) that performs rich context exchange between three decoder branches using unary, pairwise, and ternary relations of human, object, and interaction tokens. The proposed method learns comprehensive relational contexts for discovering HOI instances, achieving state-of-the-art performance on two standard benchmarks for HOI detection, HICO-DET and V-COCO.
Kinship recognition aims to determine whether the subjects in two facial images are kin or non-kin, which is an emerging and challenging problem. However, most previous methods focus on heuristic designs without considering the spatial correlation between face images. In this paper, we aim to learn discriminative kinship representations embedded with the relation information between face components (e.g., eyes, nose, etc.). To achieve this goal, we propose the Face Componential Relation Network, which learns the relationship between face components among images with a cross-attention mechanism, which automatically learns the important facial regions for kinship recognition. Moreover, we propose Face Componential Relation Network (FaCoRNet), which adapts the loss function by the guidance from cross-attention to learn more discriminative feature representations. The proposed FaCoRNet outperforms previous state-of-the-art methods by large margins for the largest public kinship recognition FIW benchmark.
The relation modeling between actors and scene context advances video action detection where the correlation of multiple actors makes their action recognition challenging. Existing studies model each actor and scene relation to improve action recognition. However, the scene variations and background interference limit the effectiveness of this relation modeling. In this paper, we propose to select actor-related scene context, rather than directly leverage raw video scenario, to improve relation modeling. We develop a Cycle Actor-Context Relation network (CycleACR) where there is a symmetric graph that models the actor and context relations in a bidirectional form. Our CycleACR consists of the Actor-to-Context Reorganization (A2C-R) that collects actor features for context feature reorganizations, and the Context-to-Actor Enhancement (C2A-E) that dynamically utilizes reorganized context features for actor feature enhancement. Compared to existing designs that focus on C2A-E, our CycleACR introduces A2C-R for a more effective relation modeling. This modeling advances our CycleACR to achieve state-of-the-art performance on two popular action detection datasets (i.e., AVA and UCF101-24). We also provide ablation studies and visualizations as well to show how our cycle actor-context relation modeling improves video action detection. Code is available at https://github.com/MCG-NJU/CycleACR.
3D scene understanding, e.g., point cloud semantic and instance segmentation, often requires large-scale annotated training data, but clearly, point-wise labels are too tedious to prepare. While some recent methods propose to train a 3D network with small percentages of point labels, we take the approach to an extreme and propose ``One Thing One Click,'' meaning that the annotator only needs to label one point per object. To leverage these extremely sparse labels in network training, we design a novel self-training approach, in which we iteratively conduct the training and label propagation, facilitated by a graph propagation module. Also, we adopt a relation network to generate the per-category prototype to enhance the pseudo label quality and guide the iterative training. Besides, our model can be compatible to 3D instance segmentation equipped with a point-clustering strategy. Experimental results on both ScanNet-v2 and S3DIS show that our self-training approach, with extremely-sparse annotations, outperforms all existing weakly supervised methods for 3D semantic and instance segmentation by a large margin, and our results are also comparable to those of the fully supervised counterparts. Codes and models are available at https://github.com/liuzhengzhe/One-Thing-One-Click.