Kai Liu, Ze Chen, Zhihang Fu, Wei Zhang, Rongxin Jiang, Fan Zhou, Yaowu Chen, Yue Wu, Jieping Ye. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Multi-label zero-shot learning (ML-ZSL) strives to recognize all objects in an image, regardless of whether they are present in the training data. Recent methods incorporate an attention mechanism to locate labels in the image and generate class-specific semantic information. However, the attention mechanism built on visual features treats label embeddings equally in the prediction score, leading to severe semantic ambiguity. This study focuses on efficiently utilizing semantic information in the attention mechanism. We propose a contrastive label-based attention method (CLA) to associate each label with the most relevant image regions. Specifically, our label-based attention, guided by the latent label embedding, captures discriminative image details. To distinguish region-wise correlations, we implement a region-level contrastive loss. In addition, we utilize a global feature alignment module to identify labels with general information. Extensive experiments on two benchmarks, NUS-WIDE and Open Images, demonstrate that our CLA outperforms the state-of-the-art methods. Especially under the ZSL setting, our method achieves 2.0% improvements in mean Average Precision (mAP) for NUS-WIDE and 4.0% for Open Images compared with recent methods.

Enlarging input images is a straightforward and effective approach to promote small object detection. However, simple image enlargement is significantly expensive on both computations and GPU memory. In fact, small objects are usually sparsely distributed and locally clustered. Therefore, massive feature extraction computations are wasted on the non-target background area of images. Recent works have tried to pick out target-containing regions using an extra network and perform conventional object detection, but the newly introduced computation limits their final performance. In this paper, we propose to reuse the detector's backbone to conduct feature-level object-seeking and patch-slicing, which can avoid redundant feature extraction and reduce the computation cost. Incorporating with a sparse detection head, we are able to detect small objects on high-resolution inputs (e.g., 1080P or larger) for superior performance. The resulting Efficient Small Object Detection (ESOD) approach is a generic framework, which can be applied to both CNN- and ViT-based detectors to save the computation and GPU memory costs. Extensive experiments demonstrate the efficacy and efficiency of our method. In particular, our method consistently surpasses the SOTA detectors by a large margin (e.g., 8% gains on AP) on the representative VisDrone, UAVDT, and TinyPerson datasets.

Multi-view Clustering (MVC) has achieved significant progress, with many efforts dedicated to learn knowledge from multiple views. However, most existing methods are either not applicable or require additional steps for incomplete MVC. Such a limitation results in poor-quality clustering performance and poor missing view adaptation. Besides, noise or outliers might significantly degrade the overall clustering performance, which are not handled well by most existing methods. In this paper, we propose a novel unified framework for incomplete and complete MVC named self-learning symmetric multi-view probabilistic clustering (SLS-MPC). SLS-MPC proposes a novel symmetric multi-view probability estimation and equivalently transforms multi-view pairwise posterior matching probability into composition of each view's individual distribution, which tolerates data missing and might extend to any number of views. Then, SLS-MPC proposes a novel self-learning probability function without any prior knowledge and hyper-parameters to learn each view's individual distribution. Next, graph-context-aware refinement with path propagation and co-neighbor propagation is used to refine pairwise probability, which alleviates the impact of noise and outliers. Finally, SLS-MPC proposes a probabilistic clustering algorithm to adjust clustering assignments by maximizing the joint probability iteratively without category information. Extensive experiments on multiple benchmarks show that SLS-MPC outperforms previous state-of-the-art methods.

Detecting and rejecting unknown out-of-distribution (OOD) samples is critical for deployed neural networks to void unreliable predictions. In real-world scenarios, however, the efficacy of existing OOD detection methods is often impeded by the inherent imbalance of in-distribution (ID) data, which causes significant performance decline. Through statistical observations, we have identified two common challenges faced by different OOD detectors: misidentifying tail class ID samples as OOD, while erroneously predicting OOD samples as head class from ID. To explain this phenomenon, we introduce a generalized statistical framework, termed ImOOD, to formulate the OOD detection problem on imbalanced data distribution. Consequently, the theoretical analysis reveals that there exists a class-aware bias item between balanced and imbalanced OOD detection, which contributes to the performance gap. Building upon this finding, we present a unified training-time regularization technique to mitigate the bias and boost imbalanced OOD detectors across architecture designs. Our theoretically grounded method translates into consistent improvements on the representative CIFAR10-LT, CIFAR100-LT, and ImageNet-LT benchmarks against several state-of-the-art OOD detection ap- proaches. Code is available at https://github.com/alibaba/imood.

When reading long-form text, human cognition is complex and structurized. While large language models (LLMs) process input contexts through a causal and sequential perspective, this approach can potentially limit their ability to handle intricate and complex inputs effectively. To enhance LLM’s cognition capability, this paper presents a novel concept of context structurization. Specifically, we transform the plain, unordered contextual sentences into well-ordered and hierarchically structurized elements. By doing so, LLMs can better grasp intricate and extended contexts through precise attention and information-seeking along the organized structures. Extensive evaluations are conducted across various model architectures and sizes (including a series of auto-regressive LLMs as well as BERT-like masking models) on a diverse set of NLP tasks (e.g., context-based question-answering, exhaustive hallucination evaluation, and passage-level dense retrieval). Empirical results show consistent and significant performance gains afforded by a single-round structurization. In particular, we boost the open-sourced LLaMA2-70B model to achieve comparable performance against GPT-3.5-Turbo as the halluci- nation evaluator. Besides, we show the feasibility of distilling advanced LLMs’ language processing abilities to a smaller yet effective StruXGPT-7B to execute structurization, addressing the practicality of our approach. Code is available at https://github.com/alibaba/struxgpt.

Abstract Despite the promising progress that has been made, large-scale clustering tasks still face various challenges: (i) high time and space complexity in K-nearest neighbors (KNN), which is often overlooked by most methods, and (ii) low recall rate caused by simply splitting the dataset. In this paper, we propose a novel framework for large-scale clustering tasks named large-scale clustering via recall KNN and subgraph segmentation (LS-RKSS) to perform faster clustering with guaranteed clustering performance, which embraces the ability of handling large-scale data up to 100 million using a single T4 GPU with less than 10% of the running time. We propose recall KNN (RKNN) and subgraph segmentation (SS) to effectively address the primary challenges in large-scale clustering tasks. Firstly, the recall KNN is proposed to perform efficient similarity search among dense vectors with lower time and space complexity compared to traditional exact search methods of KNN. Then, the subgraph segmentation is proposed to split the whole dataset into multiple subgraphs based on the recall KNN. Given the recall rate of RKNN based on traditional exact search methods, it is theoretically proved that dividing the dataset into multiple subgraphs using recall KNN and subgraph segmentation is a more reasonable and effective approach. Finally, clusters are generated independently on each subgraph, and the final clustering result is obtained by combining the results of all subgraphs. Extensive experiments demonstrate that LS-RKSS outperforms previous large-scale clustering methods in both effectiveness and efficiency.

This letter proposes a novel method for effectively utilizing multiple teachers in feature-based knowledge distillation. Our method involves a multi-teacher feature ensemble module for generating a robust feature ensemble and a student-teacher mapping module for bridging the student feature and ensemble feature. In addition, we utilize separate optimization, where the student's feature extractor is optimized under distillation supervision while its classifier is obtained through classifier reconstruction. We evaluate our method on the CIFAR-100, ImageNet and MS-COCO datasets, and the experimental results demonstrate its effectiveness.

Without manually annotated identities, unsupervised multi-object trackers are inferior to learning reliable feature embeddings. It causes the similarity-based inter-frame association stage also be error-prone, where an uncertainty problem arises. The frame-by-frame accumulated uncertainty prevents trackers from learning the consistent feature embedding against time variation. To avoid this uncertainty problem, recent self-supervised techniques are adopted, whereas they failed to capture temporal relations. The inter-frame uncertainty still exists. In fact, this paper argues that though the uncertainty problem is inevitable, it is possible to leverage the uncertainty itself to improve the learned consistency in turn. Specifically, an uncertainty-based metric is developed to verify and rectify the risky associations. The resulting accurate pseudo-tracklets boost learning the feature consistency. And accurate tracklets can incorporate temporal information into spatial transformation. This paper proposes a tracklet-guided augmentation strategy to simulate the tracklet’s motion, which adopts a hierarchical uncertainty-based sampling mechanism for hard sample mining. The ultimate unsupervised MOT framework, namely U2MOT, is proven effective on MOT-Challenges and VisDrone-MOT benchmark. U2MOT achieves a SOTA performance among the published supervised and unsupervised trackers.

The key to OOD detection has two aspects: generalized feature representation and precise category description. Recently, vision-language models such as CLIP provide significant advances in both two issues, but constructing precise category descriptions is still in its infancy due to the absence of unseen categories. This work introduces two hierarchical contexts, namely perceptual context and spurious context, to carefully describe the precise category boundary through automatic prompt tuning. Specifically, perceptual contexts perceive the inter-category difference (e.g., cats vs apples) for current classification tasks, while spurious contexts further identify spurious (similar but exactly not) OOD samples for every single category (e.g., cats vs panthers, apples vs peaches). The two contexts hierarchically construct the precise description for a certain category, which is, first roughly classifying a sample to the predicted category and then delicately identifying whether it is truly an ID sample or actually OOD. Moreover, the precise descriptions for those categories within the vision-language framework present a novel application: CATegory-EXtensible OOD detection (CATEX). One can efficiently extend the set of recognizable categories by simply merging the hierarchical contexts learned under different sub-task settings. And extensive experiments are conducted to demonstrate CATEX’s effectiveness, robustness, and category-extensibility. For instance, CATEX consistently surpasses the rivals by a large margin with several protocols on the challenging ImageNet-1K dataset. In addition, we offer new insights on how to efficiently scale up the prompt engineering in vision-language models to recognize thousands of object categories, as well as how to incorporate large language models (like GPT-3) to boost zero-shot applications.

Despite the promising progress having been made, the two challenges of multi-view clustering (MVC) are still waiting for better solutions: i) Most existing methods are either not qualified or require additional steps for incomplete multi-view clustering and ii) noise or outliers might significantly degrade the overall clustering performance. In this paper, we propose a novel unified framework for incomplete and complete MVC named multi-view probabilistic clustering (MPC). MPC equivalently transforms multi-view pairwise posterior matching probability into composition of each view's individual distribution, which tolerates data missing and might extend to any number of views. Then graph-context-aware refinement with path propagation and co-neighbor propagation is used to refine pairwise probability, which alleviates the impact of noise and outliers. Finally, MPC also equivalently transforms probabilistic clustering's objective to avoid complete pairwise computation and adjusts clustering assignments by maximizing joint probability iteratively. Extensive experiments on multiple benchmarks for incomplete and complete MVC show that MPC significantly outperforms previous state-of-the-art methods in both effectiveness and efficiency.

Recently, some works found an interesting phenomenon that adversarially robust classifiers can generate good images comparable to generative models. We investigate this phenomenon from an energy perspective and provide a novel explanation. We reformulate adversarial example generation, adversarial training, and image generation in terms of an energy function. We find that adversarial training contributes to obtaining an energy function that is flat and has low energy around the real data, which is the key for generative capability. Based on our new understanding, we further propose a better adversarial training method, Joint Energy Adversarial Training (JEAT), which can generate high-quality images and achieve new state-of-the-art robustness under a wide range of attacks. The Inception Score of the images (CIFAR-10) generated by JEAT is 8.80, much better than original robust classifiers (7.50). In particular, we find that the robustness of JEAT is better than other hybrid models.

Based on the framework of multiple instance learning (MIL), tremendous works have promoted the advances of weakly supervised object detection (WSOD). However, most MIL-based methods tend to localize instances to their discriminative parts instead of the whole content. In this paper, we propose a spatial likelihood voting (SLV) module to converge the proposal localizing process without any bounding box annotations. Specifically, all region proposals in a given image play the role of voters every iteration during training, voting for the likelihood of each category in spatial dimensions. After dilating alignment on the area with large likelihood values, the voting results are regularized as bounding boxes, being used for the final classification and localization. Based on SLV, we further propose an end-to-end training framework for multi-task learning. The classification and localization tasks promote each other, which further improves the detection performance. Extensive experiments on the PASCAL VOC 2007 and 2012 datasets demonstrate the superior performance of SLV.

Today's scene graph generation (SGG) task is largely limited in realistic scenarios, mainly due to the extremely long-tailed bias of predicate annotation distribution. Thus, tackling the class imbalance trouble of SGG is critical and challenging. In this paper, we first discover that when predicate labels have strong correlation with each other, prevalent re-balancing strategies (e.g., re-sampling and re-weighting) will give rise to either over-fitting the tail data (e.g., bench sitting on sidewalk rather than on), or still suffering the adverse effect from the original uneven distribution (e.g., aggregating varied parked on/standing on/sitting on into on). We argue the principal reason is that re-balancing strategies are sensitive to the frequencies of predicates yet blind to their relatedness, which may play a more important role to promote the learning of predicate features. Therefore, we propose a novel Predicate-Correlation Perception Learning (PCPL for short) scheme to adaptively seek out appropriate loss weights by directly perceiving and utilizing the correlation among predicate classes. Moreover, our PCPL framework is further equipped with a graph encoder module to better extract context features. Extensive experiments on the benchmark VG150 dataset show that the proposed PCPL performs markedly better on tail classes while well-preserving the performance on head ones, which significantly outperforms previous state-of-the-art methods.

This paper presents a compact neural network for video frame interpolation using phase and amplitude feature pyramids. We design a set of one-dimensional separable complex Gabor filters to extract phase and amplitude feature pyramids for each input image, which is efficient and effective for motion representation. The pyramids are fused and fed into a decoder network to estimate bi-directional optical flow. The interpolated frame is refined by a context-aware synthesis module. We train our model on quintets of frames using motion linear regularization. The proposed network contains much fewer parameters than state-of-the-art approaches. The experiments show that our method outperforms the competing methods. Moreover, our method achieves marked visual improvement in the challenging scenario with lighting changes.

In this paper, we present novel sharp attention networks by adaptively sampling feature maps from convolutional neural networks for person re-identification (re-ID) problems. Due to the introduction of sampling-based attention models, the proposed approach can adaptively generate sharper attention-aware feature masks. This greatly differs from the gating-based attention mechanism that relies on soft gating functions to select the relevant features for person re-ID. In contrast, the proposed sampling-based attention mechanism allows us to effectively trim irrelevant features by enforcing the resultant feature masks to focus on the most discriminative features. It can produce sharper attentions that is more assertive in localizing subtle features relevant to re-identifying people across cameras. For this purpose, a differentiable Gumbel-Softmax sampler is employed to approximate the Bernoulli sampling to train the sharp attention networks. Extensive experimental evaluations demonstrate the superiority of this new sharp attention model for person re-ID over other related existing, published state-of-the-art works on three challenging benchmarks, including CUHK03, Market-1501, and DukeMTMC-reID.

Detecting objects in surveillance videos is an important problem due to its wide applications in traffic control and public security. Existing methods tend to face performance degradation because of false positive or misalignment problems. We propose a novel framework, namely, Foreground Gating and Background Refining Network (FG-BR Net), for surveillance object detection (SOD). To reduce false positives in background regions, which is a critical problem in SOD, we introduce a new module that first subtracts the background of a video sequence and then generates high-quality region proposals. Unlike previous background subtraction methods that may wrongly remove the static foreground objects in a frame, a feedback connection from detection results to background subtraction process is proposed in our model to distill both static and moving objects in surveillance videos. Furthermore, we introduce another module, namely, the background refining stage, to refine the detection results with more accurate localizations. Pairwise non-local operations are adopted to cope with the misalignments between the features of original and background frames. Extensive experiments on real-world traffic surveillance benchmarks demonstrate the competitive performance of the proposed FG-BR Net. In particular, FG-BR Net ranks on the top among all the methods on hard and sunny subsets of the UA-DETRAC detection dataset, without any bells and whistles.

In this article, we propose a novel deep Siamese architecture based on a convolutional neural network (CNN) and multi-level similarity perception for the person re-identification (re-ID) problem. According to the distinct characteristics of diverse feature maps, we effectively apply different similarity constraints to both low-level and high-level feature maps during training stage. Due to the introduction of appropriate similarity comparison mechanisms at different levels, the proposed approach can adaptively learn discriminative local and global feature representations, respectively, while the former is more sensitive in localizing part-level prominent patterns relevant to re-identifying people across cameras. Meanwhile, a novel strong activation pooling strategy is utilized on the last convolutional layer for abstract local-feature aggregation to pursue more representative feature representations. Based on this, we propose final feature embedding by simultaneously encoding original global features and discriminative local features. In addition, our framework has two other benefits: First, classification constraints can be easily incorporated into the framework, forming a unified multi-task network with similarity constraints. Second, as similarity-comparable information has been encoded in the network’s learning parameters via back-propagation, pairwise input is not necessary at test time. That means we can extract features of each gallery image and build an index in an off-line manner, which is essential for large-scale real-world applications. Experimental results on multiple challenging benchmarks demonstrate that our method achieves splendid performance compared with the current state-of-the-art approaches.

Most multi-scale detectors face a challenge of small-size false positives due to the inadequacy of low-level features, which have small receptive field sizes and weak semantic capabilities. This paper demonstrates independent predictions from different feature layers on the same region is beneficial for reducing false positives. We propose a novel light-weight previewer block, which previews the objectness probability for the potential regression region of each prior box, using the stronger features with larger receptive fields and more contextual information for better predictions. This previewer block is generic and can be easily implemented in multi-scale detectors, such as SSD, RFBNet and MS-CNN. Extensive experiments are conducted on PASCAL VOC and KITTI pedestrian benchmark to show the superiority of the proposed method.

Person re-identification (re-ID), which aims at spotting a person of interest across multiple camera views, has gained more and more attention in computer vision community. In this paper, we propose a novel deep Siamese architecture based on convolutional neural network (CNN) and multi-level similarity perception. According to the distinct characteristics of diverse feature maps, we effectively apply different similarity constraints to both low-level and high-level feature maps, during training stage. Therefore, our network can efficiently learn discriminative feature representations at different levels, which significantly improves the re-ID performance. Besides, our framework has two additional benefits. Firstly, classification constraints can be easily incorporated into the framework, forming a unified multi-task network with similarity constraints. Secondly, as similarity comparable information has been encoded in the network's learning parameters via back-propagation, pairwise input is not necessary at test time. That means we can extract features of each gallery image and build index in an off-line manner, which is essential for large-scale real-world applications. Experimental results on multiple challenging benchmarks demonstrate that our method achieves splendid performance compared with the current state-of-the-art approaches.