Xiangyu Wu

Dingqi Zhang, Antonio Loquercio, Xiangyu Wu et al. · berkeley

This paper proposes an adaptive near-hover position controller for quadcopters, which can be deployed to quadcopters of very different mass, size and motor constants, and also shows rapid adaptation to unknown disturbances during runtime. The core algorithmic idea is to learn a single policy that can adapt online at test time not only to the disturbances applied to the drone, but also to the robot dynamics and hardware in the same framework. We achieve this by training a neural network to estimate a latent representation of the robot and environment parameters, which is used to condition the behaviour of the controller, also represented as a neural network. We train both networks exclusively in simulation with the goal of flying the quadcopters to goal positions and avoiding crashes to the ground. We directly deploy the same controller trained in the simulation without any modifications on two quadcopters in the real world with differences in mass, size, motors, and propellers with mass differing by 4.5 times. In addition, we show rapid adaptation to sudden and large disturbances up to one-third of the mass of the quadcopters. We perform an extensive evaluation in both simulation and the physical world, where we outperform a state-of-the-art learning-based adaptive controller and a traditional PID controller specifically tuned to each platform individually. Video results can be found at https://youtu.be/U-c-LbTfvoA.

OneRec Team, Biao Yang, Boyang Ding et al.

Generative recommendation models in the OneRec family have been widely deployed in many real-world services, such as short-video, live-streaming, advertising, and e-commerce. However, these generative models can only benefit from the scaling advantage, while their reasoning ability is hard to activate, since we cannot construct meaningful Chain-of-Thought (CoT) sequences consisting of itemic tokens only. Inspired by the success of the reasoning-style ``think before answer'' paradigm in the LLM field, we conduct preliminary studies (i.e., OneRec-Think, OpenOneRec) to explore reasoning capability in generative recommendation. Nevertheless, we notice an unexpected phenomenon: the thinking mode does not show advantages over the non-thinking mode. Drawing insights from recent findings on CoT robustness in multi-modal language models, we argue that effective reasoning in recommendation rests on two factors: perception, the ability to ground itemic tokens in their underlying language semantics, and cognition, the ability to reorganize a user's behavior sequence into coherent latent interest points. We therefore propose OneReason, which includes: (1) strong itemic token perception in pre-training, (2) a three-level cognition-enhanced CoT format for recommendation tasks in SFT, and (3) a specialize-then-unify training recipe in RL to enhance the thinking ability.

Taehoon Kim, Pyunghwan Ahn, Sangyun Kim et al. · nvidia, utoronto

In this report, we introduce NICE (New frontiers for zero-shot Image Captioning Evaluation) project and share the results and outcomes of 2023 challenge. This project is designed to challenge the computer vision community to develop robust image captioning models that advance the state-of-the-art both in terms of accuracy and fairness. Through the challenge, the image captioning models were tested using a new evaluation dataset that includes a large variety of visual concepts from many domains. There was no specific training data provided for the challenge, and therefore the challenge entries were required to adapt to new types of image descriptions that had not been seen during training. This report includes information on the newly proposed NICE dataset, evaluation methods, challenge results, and technical details of top-ranking entries. We expect that the outcomes of the challenge will contribute to the improvement of AI models on various vision-language tasks.

Le Zhang, Xiaolan Zhu, Yuchen Wang et al.

As live streaming services grow, many platforms offer short videos and live streams to meet diverse needs. Short videos carry substantial traffic and rich behavior signals, whereas live streaming is a core conversion scenario with sparse behavior data, making cold start severe. Transferring user interests from short videos to live streaming recommendation can alleviate these issues. Meanwhile, short videos and live streams are complex multimodal items, and integrating multimodal signals improves recommendation performance. Although Multimodal Large Language Models (MLLMs) show strong multimodal understanding and reasoning, their application to cross-domain recommendation remains underexplored. To this end, we propose Reasoning-Guided Cross-Domain Representation Learning (RGCD-Rep), a reasoning-guided framework for cross-domain recommendation from short videos to live streams. RGCD-Rep introduces MLLM reasoning resource-efficiently and learns transferable item representations guided by behavioral collaboration via two-stage training. First, reasoning-aware distillation lets a frozen teacher MLLM generate structured cross-domain reasoning knowledge and distills it into a lightweight student MLLM. Second, transferability-guided cross-domain representation learning decomposes item representations into transferable and domain residual representations. The resulting representations are computed offline and integrated into downstream retrieval tasks, enabling low-cost industrial deployment. Extensive offline experiments demonstrate RGCD-Rep's superiority. After deployment in Kuaishou's live streaming recommendation system, A/B tests show significant gains across multiple core business metrics, confirming its effectiveness and practicality in real industrial scenarios. RGCD-Rep is fully deployed and serves over 400 million users daily.

Zhiwen Wang, Chen Shen, Yin Xu et al.

Microgrids are resources that can be used to restore critical loads after a natural disaster, enhancing resilience of a distribution network. To deal with the stochastic nature of intermittent energy resources, such as wind turbines (WTs) and photovoltaics (PVs), many methods rely on forecast information. However, some microgrids may not be equipped with power forecasting tools. To fill this gap, a risk-limiting strategy based on measurements is proposed. Gaussian mixture model (GMM) is used to represent a prior joint probability density function (PDF) of power outputs of WTs and PVs over multiple periods. As time rolls forward, the distribution of WT/PV generation is updated based the latest measurement data in a recursive manner. The updated distribution is used as an input for the risk-limiting load restoration problem, enabling an equivalent transformation of the original chance constrained problem into a mixed integer linear programming (MILP). Simulation cases on a distribution system with three microgrids demonstrate the effectiveness of the proposed method. Results also indicate that networked microgrids have better uncertainty management capabilities than stand-alone microgrids.

Jiaxin Deng, Dong Shen, Shiyao Wang et al.

In recent years, live streaming platforms have gained immense popularity as they allow users to broadcast their videos and interact in real-time with hosts and peers. Due to the dynamic changes of live content, accurate recommendation models are crucial for enhancing user experience. However, most previous works treat the live as a whole item and explore the Click-through-Rate (CTR) prediction framework on item-level, neglecting that the dynamic changes that occur even within the same live room. In this paper, we proposed a ContentCTR model that leverages multimodal transformer for frame-level CTR prediction. First, we present an end-to-end framework that can make full use of multimodal information, including visual frames, audio, and comments, to identify the most attractive live frames. Second, to prevent the model from collapsing into a mediocre solution, a novel pairwise loss function with first-order difference constraints is proposed to utilize the contrastive information existing in the highlight and non-highlight frames. Additionally, we design a temporal text-video alignment module based on Dynamic Time Warping to eliminate noise caused by the ambiguity and non-sequential alignment of visual and textual information. We conduct extensive experiments on both real-world scenarios and public datasets, and our ContentCTR model outperforms traditional recommendation models in capturing real-time content changes. Moreover, we deploy the proposed method on our company platform, and the results of online A/B testing further validate its practical significance.

Jiaxin Deng, Dong Shen, Haojie Pan et al.

Video understanding is an important task in short video business platforms and it has a wide application in video recommendation and classification. Most of the existing video understanding works only focus on the information that appeared within the video content, including the video frames, audio and text. However, introducing common sense knowledge from the external Knowledge Graph (KG) dataset is essential for video understanding when referring to the content which is less relevant to the video. Owing to the lack of video knowledge graph dataset, the work which integrates video understanding and KG is rare. In this paper, we propose a heterogeneous dataset that contains the multi-modal video entity and fruitful common sense relations. This dataset also provides multiple novel video inference tasks like the Video-Relation-Tag (VRT) and Video-Relation-Video (VRV) tasks. Furthermore, based on this dataset, we propose an end-to-end model that jointly optimizes the video understanding objective with knowledge graph embedding, which can not only better inject factual knowledge into video understanding but also generate effective multi-modal entity embedding for KG. Comprehensive experiments indicate that combining video understanding embedding with factual knowledge benefits the content-based video retrieval performance. Moreover, it also helps the model generate better knowledge graph embedding which outperforms traditional KGE-based methods on VRT and VRV tasks with at least 42.36% and 17.73% improvement in HITS@10.

Xing Cheng, Xiangyu Wu, Dong Shen et al.

Video grounding aims to locate the timestamps best matching the query description within an untrimmed video. Prevalent methods can be divided into moment-level and clip-level frameworks. Moment-level approaches directly predict the probability of each transient moment to be the boundary in a global perspective, and they usually perform better in coarse grounding. On the other hand, clip-level ones aggregate the moments in different time windows into proposals and then deduce the most similar one, leading to its advantage in fine-grained grounding. In this paper, we propose a multi-level unified framework to enhance performance by leveraging the merits of both moment-level and clip-level methods. Moreover, a novel generation-guided paradigm in both levels is adopted. It introduces a multi-modal generator to produce the implicit boundary feature and clip feature, later regarded as queries to calculate the boundary scores by a discriminator. The generation-guided solution enhances video grounding from a two-unique-modals' match task to a cross-modal attention task, which steps out of the previous framework and obtains notable gains. The proposed Generation-guided Multi-level Unified network (GMU) surpasses previous methods and reaches State-Of-The-Art on various benchmarks with disparate features, e.g., Charades-STA, ActivityNet captions.

Zhuoqun Li, Boxi Cao, Guiping Jiang et al.

Metaphorical videos are prevalent across various real-world scenarios to convey complex ideas, and understanding them typically requires high-order cognitive capabilities. The lack of systematic studies on metaphorical video understanding not only constrains the real-world applicability of MLLMs but also impedes the thorough assessment of their high-order cognitive capabilities. To bridge this gap, we propose MetaphorVU-Bench, the first systematic and comprehensive benchmark dedicated to metaphorical video understanding. Through experiments, we find current MLLMs struggle with accurate metaphorical video understanding, lagging far behind human level, primarily due to defective cross-domain mapping. Motivated by this finding, we construct a metaphor knowledge graph as mapping augmentation and propose MetaphorBoost, an inference-time enhancement framework achieving consistent performance improvement. Our benchmark, analysis, and method provide useful insights and a foundation for future research on advancing MLLMs.

Feng Yu, Xiangyu Wu, Yang Yang et al.

Multimodal learning integrates data from diverse sensors to effectively harness information from different modalities. However, recent studies reveal that joint learning often overfits certain modalities while neglecting others, leading to performance inferior to that of unimodal learning. Although previous efforts have sought to balance modal contributions or combine joint and unimodal learning, thereby mitigating the degradation of weaker modalities with promising outcomes, few have examined the relationship between joint and unimodal learning from an information-theoretic perspective. In this paper, we theoretically analyze modality competition and propose a method for multimodal classification by maximizing the total correlation between multimodal features and labels. By maximizing this objective, our approach alleviates modality competition while capturing inter-modal interactions via feature alignment. Building on Mutual Information Neural Estimation (MINE), we introduce Total Correlation Neural Estimation (TCNE) to derive a lower bound for total correlation. Subsequently, we present TCMax, a hyperparameter-free loss function that maximizes total correlation through variational bound optimization. Extensive experiments demonstrate that TCMax outperforms state-of-the-art joint and unimodal learning approaches. Our code is available at https://github.com/hubaak/TCMax.

Xiangyu Wu, Dongming Jiang, Feng Yu et al.

Mainstream Test-Time Adaptation (TTA) methods for adapting vision-language models, e.g., CLIP, typically rely on Shannon Entropy (SE) at test time to measure prediction uncertainty and inconsistency. However, since CLIP has a built-in bias from pretraining on highly imbalanced web-crawled data, SE inevitably results in producing biased estimates of uncertainty entropy. To address this issue, we notably find and demonstrate that Tsallis Entropy (TE), a generalized form of SE, is naturally suited for characterizing biased distributions by introducing a non-extensive parameter q, with the performance of SE serving as a lower bound for TE. Building upon this, we generalize TE into Adaptive Debiasing Tsallis Entropy (ADTE) for TTA, customizing a class-specific parameter q^l derived by normalizing the estimated label bias from continuously incoming test instances, for each category. This adaptive approach allows ADTE to accurately select high-confidence views and seamlessly integrate with a label adjustment strategy to enhance adaptation, without introducing distribution-specific hyperparameter tuning. Besides, our investigation reveals that both TE and ADTE can serve as direct, advanced alternatives to SE in TTA, without any other modifications. Experimental results show that ADTE outperforms state-of-the-art methods on ImageNet and its five variants, and achieves the highest average performance on 10 cross-domain benchmarks, regardless of the model architecture or text prompts used. Our code is available at https://github.com/Jinx630/ADTE.

Yang Yang, Zhongtian Fu, Xiangyu Wu et al.

Current vision-language retrieval aims to perform cross-modal instance search, in which the core idea is to learn the consistent visionlanguage representations. Although the performance of cross-modal retrieval has greatly improved with the development of deep models, we unfortunately find that traditional hard consistency may destroy the original relationships among single-modal instances, leading the performance degradation for single-modal retrieval. To address this challenge, in this paper, we experimentally observe that the vision-language divergence may cause the existence of strong and weak modalities, and the hard cross-modal consistency cannot guarantee that strong modal instances' relationships are not affected by weak modality, resulting in the strong modal instances' relationships perturbed despite learned consistent representations.To this end, we propose a novel and directly Coordinated VisionLanguage Retrieval method (dubbed CoVLR), which aims to study and alleviate the desynchrony problem between the cross-modal alignment and single-modal cluster-preserving tasks. CoVLR addresses this challenge by developing an effective meta-optimization based strategy, in which the cross-modal consistency objective and the intra-modal relation preserving objective are acted as the meta-train and meta-test tasks, thereby CoVLR encourages both tasks to be optimized in a coordinated way. Consequently, we can simultaneously insure cross-modal consistency and intra-modal structure. Experiments on different datasets validate CoVLR can improve single-modal retrieval accuracy whilst preserving crossmodal retrieval capacity compared with the baselines.

Xiangyu Wu, Qing-Yuan Jiang, Yang Yang et al.

The recent introduction of prompt tuning based on pre-trained vision-language models has dramatically improved the performance of multi-label image classification. However, some existing strategies that have been explored still have drawbacks, i.e., either exploiting massive labeled visual data at a high cost or using text data only for text prompt tuning and thus failing to learn the diversity of visual knowledge. Hence, the application scenarios of these methods are limited. In this paper, we propose a pseudo-visual prompt~(PVP) module for implicit visual prompt tuning to address this problem. Specifically, we first learn the pseudo-visual prompt for each category, mining diverse visual knowledge by the well-aligned space of pre-trained vision-language models. Then, a co-learning strategy with a dual-adapter module is designed to transfer visual knowledge from pseudo-visual prompt to text prompt, enhancing their visual representation abilities. Experimental results on VOC2007, MS-COCO, and NUSWIDE datasets demonstrate that our method can surpass state-of-the-art~(SOTA) methods across various settings for multi-label image classification tasks. The code is available at https://github.com/njustkmg/PVP.

Jiaqi Tang, Yu Xia, Yi-Feng Wu et al.

The advent of autonomous agents is transforming interactions with Graphical User Interfaces (GUIs) by employing natural language as a powerful intermediary. Despite the predominance of Supervised Fine-Tuning (SFT) methods in current GUI agents for achieving spatial localization, these methods face substantial challenges due to their limited capacity to accurately perceive positional data. Existing strategies, such as reinforcement learning, often fail to assess positional accuracy effectively, thereby restricting their utility. In response, we introduce Location Preference Optimization (LPO), a novel approach that leverages locational data to optimize interaction preferences. LPO uses information entropy to predict interaction positions by focusing on zones rich in information. Besides, it further introduces a dynamic location reward function based on physical distance, reflecting the varying importance of interaction positions. Supported by Group Relative Preference Optimization (GRPO), LPO facilitates an extensive exploration of GUI environments and significantly enhances interaction precision. Comprehensive experiments demonstrate LPO's superior performance, achieving SOTA results across both offline benchmarks and real-world online evaluations. Our code will be made publicly available soon, at https://github.com/AIDC-AI/LPO.

Lihao Liu, Yan Wang, Biao Yang et al.

Multimodal Large Language Models (MLLMs) have shown remarkable success in comprehension tasks such as visual description and visual question answering. However, their direct application to embedding-based tasks like retrieval remains challenging due to the discrepancy between output formats and optimization objectives. Previous approaches often employ contrastive fine-tuning to adapt MLLMs for retrieval, but at the cost of losing their generative capabilities. We argue that both generative and embedding tasks fundamentally rely on shared cognitive mechanisms, specifically cross-modal representation alignment and contextual comprehension. To this end, we propose CREM (Compression-driven Representation Enhanced Model), with a unified framework that enhances multimodal representations for retrieval while preserving generative ability. Specifically, we introduce a compression-based prompt design with learnable chorus tokens to aggregate multimodal semantics and a compression-driven training strategy that integrates contrastive and generative objectives through compression-aware attention. Extensive experiments demonstrate that CREM achieves state-of-the-art retrieval performance on MMEB while maintaining strong generative performance on multiple comprehension benchmarks. Our findings highlight that generative supervision can further improve the representational quality of MLLMs under the proposed compression-driven paradigm.

Zhiwen Wang, Chen Shen, Feng Liu et al.

With the fast growth of wind power penetration, power systems need additional flexibility to cope with wind power ramping. Several electricity markets have established requirements for flexible ramping capacity (FRC) reserves. This paper addresses two crucial issues that have rarely been discussed in the literature: 1) how to characterize wind power ramping under different forecast values and 2) how to achieve a reasonable trade-off between operational risks and FRC costs. Regarding the first issue, this paper proposes a concept of conditional distributions of wind power ramping, which is empirically verified by using simulation and real-world data. For the second issue, this paper develops an adjustable chance-constrained approach to optimally allocate FRC reserves. Equivalent tractable forms of the original problem are devised to improve computational efficiency. Tests carried out on a modified IEEE 118-bus system demonstrate the effectiveness and efficiency of the proposed method.

Xiangyu Wu, Feng Yu, Qing-Guo Chen et al.

Mainstream test-time adaptation (TTA) techniques endeavor to mitigate distribution shifts via entropy minimization for multi-class classification, inherently increasing the probability of the most confident class. However, when encountering multi-label instances, the primary challenge stems from the varying number of labels per image, and prioritizing only the highest probability class inevitably undermines the adaptation of other positive labels. To address this issue, we investigate TTA within multi-label scenario (ML--TTA), developing Bound Entropy Minimization (BEM) objective to simultaneously increase the confidence of multiple top predicted labels. Specifically, to determine the number of labels for each augmented view, we retrieve a paired caption with yielded textual labels for that view. These labels are allocated to both the view and caption, called weak label set and strong label set with the same size k. Following this, the proposed BEM considers the highest top-k predicted labels from view and caption as a single entity, respectively, learning both view and caption prompts concurrently. By binding top-k predicted labels, BEM overcomes the limitation of vanilla entropy minimization, which exclusively optimizes the most confident class. Across the MSCOCO, VOC, and NUSWIDE multi-label datasets, our ML--TTA framework equipped with BEM exhibits superior performance compared to the latest SOTA methods, across various model architectures, prompt initialization, and varying label scenarios. The code is available at https://github.com/Jinx630/ML-TTA.

Sayang Mu, Xiangyu Wu, Bo An

Efficient communication is critical for decentralized Multi-Robot Path Planning (MRPP), yet existing learned communication methods treat all neighboring robots equally regardless of their spatial proximity, leading to diluted attention in congested regions where coordination matters most. We propose Relation enhanced Multi Head Attention (RMHA), a communication mechanism that explicitly embeds pairwise Manhattan distances into the attention weight computation, enabling each robot to dynamically prioritize messages from spatially relevant neighbors. Combined with a distance-constrained attention mask and GRU gated message fusion, RMHA integrates seamlessly with MAPPO for stable end-to-end training. In zero-shot generalization from 8 training robots to 128 test robots on 40x40 grids, RMHA achieves approximately 75 percent success rate at 30 percent obstacle density outperforming the best baseline by over 25 percentage points. Ablation studies confirm that distance-relation encoding is the key contributor to success rate improvement in high-density environments. Index Terms-Multi-robot path planning, graph attention mechanism, multi-head attention, communication optimization, cooperative decision-making

Xiangyu Wu, Yang Yang, Shengdong Xu et al.

In this paper, we present our solution to a Multi-modal Algorithmic Reasoning Task: SMART-101 Challenge. Different from the traditional visual question-answering datasets, this challenge evaluates the abstraction, deduction, and generalization abilities of neural networks in solving visuolinguistic puzzles designed specifically for children in the 6-8 age group. We employed a divide-and-conquer approach. At the data level, inspired by the challenge paper, we categorized the whole questions into eight types and utilized the llama-2-chat model to directly generate the type for each question in a zero-shot manner. Additionally, we trained a yolov7 model on the icon45 dataset for object detection and combined it with the OCR method to recognize and locate objects and text within the images. At the model level, we utilized the BLIP-2 model and added eight adapters to the image encoder VIT-G to adaptively extract visual features for different question types. We fed the pre-constructed question templates as input and generated answers using the flan-t5-xxl decoder. Under the puzzle splits configuration, we achieved an accuracy score of 26.5 on the validation set and 24.30 on the private test set.

Xiangyu Wu, Yi Gao, Hailiang Zhang et al.

In this paper, we present our solution to the New frontiers for Zero-shot Image Captioning Challenge. Different from the traditional image captioning datasets, this challenge includes a larger new variety of visual concepts from many domains (such as COVID-19) as well as various image types (photographs, illustrations, graphics). For the data level, we collect external training data from Laion-5B, a large-scale CLIP-filtered image-text dataset. For the model level, we use OFA, a large-scale visual-language pre-training model based on handcrafted templates, to perform the image captioning task. In addition, we introduce contrastive learning to align image-text pairs to learn new visual concepts in the pre-training stage. Then, we propose a similarity-bucket strategy and incorporate this strategy into the template to force the model to generate higher quality and more matching captions. Finally, by retrieval-augmented strategy, we construct a content-rich template, containing the most relevant top-k captions from other image-text pairs, to guide the model in generating semantic-rich captions. Our method ranks first on the leaderboard, achieving 105.17 and 325.72 Cider-Score in the validation and test phase, respectively.

Jiaqi Tang, Jianmin Chen, Wei Wei et al.

Multimodal Large Language Models struggle to maintain reliable performance under extreme real-world visual degradations, which impede their practical robustness. Existing robust MLLMs predominantly rely on implicit training/adaptation that focuses solely on visual encoder generalization, suffering from limited interpretability and isolated optimization. To overcome these limitations, we propose Robust-R1, a novel framework that explicitly models visual degradations through structured reasoning chains. Our approach integrates: (i) supervised fine-tuning for degradation-aware reasoning foundations, (ii) reward-driven alignment for accurately perceiving degradation parameters, and (iii) dynamic reasoning depth scaling adapted to degradation intensity. To facilitate this approach, we introduce a specialized 11K dataset featuring realistic degradations synthesized across four critical real-world visual processing stages, each annotated with structured chains connecting degradation parameters, perceptual influence, pristine semantic reasoning chain, and conclusion. Comprehensive evaluations demonstrate state-of-the-art robustness: Robust-R1 outperforms all general and robust baselines on the real-world degradation benchmark R-Bench, while maintaining superior anti-degradation performance under multi-intensity adversarial degradations on MMMB, MMStar, and RealWorldQA.

Xiangyu Wu, Feng Yu, Yang Yang et al.

The integration of prompt tuning with multimodal learning has shown significant generalization abilities for various downstream tasks. Despite advancements, existing methods heavily depend on massive modality-specific labeled data (e.g., video, audio, and image), or are customized for a single modality. In this study, we present Text as Any-Modality by Consistent Prompt Tuning (TaAM-CPT), a scalable approach for constructing a general representation model toward unlimited modalities using solely text data. TaAM-CPT comprises modality prompt pools, text construction, and modality-aligned text encoders from pre-trained models, which allows for extending new modalities by simply adding prompt pools and modality-aligned text encoders. To harmonize the learning across different modalities, TaAM-CPT designs intra- and inter-modal learning objectives, which can capture category details within modalities while maintaining semantic consistency across different modalities. Benefiting from its scalable architecture and pre-trained models, TaAM-CPT can be seamlessly extended to accommodate unlimited modalities. Remarkably, without any modality-specific labeled data, TaAM-CPT achieves leading results on diverse datasets spanning various modalities, including video classification, image classification, and audio classification. The code is available at https://github.com/Jinx630/TaAM-CPT.

Xing Cheng, Hezheng Lin, Xiangyu Wu et al.

The task of multi-label image classification is to recognize all the object labels presented in an image. Though advancing for years, small objects, similar objects and objects with high conditional probability are still the main bottlenecks of previous convolutional neural network(CNN) based models, limited by convolutional kernels' representational capacity. Recent vision transformer networks utilize the self-attention mechanism to extract the feature of pixel granularity, which expresses richer local semantic information, while is insufficient for mining global spatial dependence. In this paper, we point out the three crucial problems that CNN-based methods encounter and explore the possibility of conducting specific transformer modules to settle them. We put forward a Multi-label Transformer architecture(MlTr) constructed with windows partitioning, in-window pixel attention, cross-window attention, particularly improving the performance of multi-label image classification tasks. The proposed MlTr shows state-of-the-art results on various prevalent multi-label datasets such as MS-COCO, Pascal-VOC, and NUS-WIDE with 88.5%, 95.8%, and 65.5% respectively. The code will be available soon at https://github.com/starmemda/MlTr/

Hezheng Lin, Xing Cheng, Xiangyu Wu et al.

Since Transformer has found widespread use in NLP, the potential of Transformer in CV has been realized and has inspired many new approaches. However, the computation required for replacing word tokens with image patches for Transformer after the tokenization of the image is vast(e.g., ViT), which bottlenecks model training and inference. In this paper, we propose a new attention mechanism in Transformer termed Cross Attention, which alternates attention inner the image patch instead of the whole image to capture local information and apply attention between image patches which are divided from single-channel feature maps capture global information. Both operations have less computation than standard self-attention in Transformer. By alternately applying attention inner patch and between patches, we implement cross attention to maintain the performance with lower computational cost and build a hierarchical network called Cross Attention Transformer(CAT) for other vision tasks. Our base model achieves state-of-the-arts on ImageNet-1K, and improves the performance of other methods on COCO and ADE20K, illustrating that our network has the potential to serve as general backbones. The code and models are available at \url{https://github.com/linhezheng19/CAT}.

Haonan Xu, Dian Chao, Xiangyu Wu et al.

Treating texts as images, combining prompts with textual labels for prompt tuning, and leveraging the alignment properties of CLIP have been successfully applied in zero-shot multi-label image recognition. Nonetheless, relying solely on textual labels to store visual information is insufficient for representing the diversity of visual objects. In this paper, we propose reversing the training process of CLIP and introducing the concept of Pseudo Visual Prompts. These prompts are initialized for each object category and pre-trained on large-scale, low-cost sentence data generated by large language models. This process mines the aligned visual information in CLIP and stores it in class-specific visual prompts. We then employ contrastive learning to transfer the stored visual information to the textual labels, enhancing their visual representation capacity. Additionally, we introduce a dual-adapter module that simultaneously leverages knowledge from the original CLIP and new learning knowledge derived from downstream datasets. Benefiting from the pseudo visual prompts, our method surpasses the state-of-the-art not only on clean annotated text data but also on pseudo text data generated by large language models.

Xiangyu Wu, Zhouyang Chi, Yang Yang et al.

In this paper, we present our solution for the WSDM2023 Toloka Visual Question Answering Challenge. Inspired by the application of multimodal pre-trained models to various downstream tasks(e.g., visual question answering, visual grounding, and cross-modal retrieval), we approached this competition as a visual grounding task, where the input is an image and a question, guiding the model to answer the question and display the answer as a bounding box on the image. We designed a three-stage solution for this task. Specifically, we used the visual-language pre-trained model OFA as the foundation. In the first stage, we constructed a large-scale synthetic dataset similar to the competition dataset and coarse-tuned the model to learn generalized semantic information. In the second stage, we treated the competition task as a visual grounding task, loaded the weights from the previous stage, and continued to fine-tune the model on the competition dataset, transferring the semantic information learned in the first stage to the competition task. Finally, we designed a bounding box matching and replacing post-processing strategy to correct the model's prediction results. Our team achieved a score of 76.342 on the final leaderboard, ranking second.

Xiangyu Wu, Jinling Xu, Longfei Huang et al.

This report introduces a solution to The task of RGB-TIR object detection from the perspective of unmanned aerial vehicles. Unlike traditional object detection methods, RGB-TIR object detection aims to utilize both RGB and TIR images for complementary information during detection. The challenges of RGB-TIR object detection from the perspective of unmanned aerial vehicles include highly complex image backgrounds, frequent changes in lighting, and uncalibrated RGB-TIR image pairs. To address these challenges at the model level, we utilized a lightweight YOLOv9 model with extended multi-level auxiliary branches that enhance the model's robustness, making it more suitable for practical applications in unmanned aerial vehicle scenarios. For image fusion in RGB-TIR detection, we incorporated a fusion module into the backbone network to fuse images at the feature level, implicitly addressing calibration issues. Our proposed method achieved an mAP score of 0.516 and 0.543 on A and B benchmarks respectively while maintaining the highest inference speed among all models.

Xiangyu Wu, Hailiang Zhang, Yang Yang et al.

In this paper, we present our champion solution to the Global Artificial Intelligence Technology Innovation Competition Track 1: Medical Imaging Diagnosis Report Generation. We select CPT-BASE as our base model for the text generation task. During the pre-training stage, we delete the mask language modeling task of CPT-BASE and instead reconstruct the vocabulary, adopting a span mask strategy and gradually increasing the number of masking ratios to perform the denoising auto-encoder pre-training task. In the fine-tuning stage, we design iterative retrieval augmentation and noise-aware similarity bucket prompt strategies. The retrieval augmentation constructs a mini-knowledge base, enriching the input information of the model, while the similarity bucket further perceives the noise information within the mini-knowledge base, guiding the model to generate higher-quality diagnostic reports based on the similarity prompts. Surprisingly, our single model has achieved a score of 2.321 on leaderboard A, and the multiple model fusion scores are 2.362 and 2.320 on the A and B leaderboards respectively, securing first place in the rankings.

Yurui Huang, Yang Yang, Shou Chen et al.

In this paper, we propose a solution for improving the quality of temporal sound localization. We employ a multimodal fusion approach to combine visual and audio features. High-quality visual features are extracted using a state-of-the-art self-supervised pre-training network, resulting in efficient video feature representations. At the same time, audio features serve as complementary information to help the model better localize the start and end of sounds. The fused features are trained in a multi-scale Transformer for training. In the final test dataset, we achieved a mean average precision (mAP) of 0.33, obtaining the second-best performance in this track.

Kwai Keye Team, Biao Yang, Bin Wen et al.

While Multimodal Large Language Models (MLLMs) demonstrate remarkable capabilities on static images, they often fall short in comprehending dynamic, information-dense short-form videos, a dominant medium in today's digital landscape. To bridge this gap, we introduce \textbf{Kwai Keye-VL}, an 8-billion-parameter multimodal foundation model engineered for leading-edge performance in short-video understanding while maintaining robust general-purpose vision-language abilities. The development of Keye-VL rests on two core pillars: a massive, high-quality dataset exceeding 600 billion tokens with a strong emphasis on video, and an innovative training recipe. This recipe features a four-stage pre-training process for solid vision-language alignment, followed by a meticulous two-phase post-training process. The first post-training stage enhances foundational capabilities like instruction following, while the second phase focuses on stimulating advanced reasoning. In this second phase, a key innovation is our five-mode ``cold-start'' data mixture, which includes ``thinking'', ``non-thinking'', ``auto-think'', ``think with image'', and high-quality video data. This mixture teaches the model to decide when and how to reason. Subsequent reinforcement learning (RL) and alignment steps further enhance these reasoning capabilities and correct abnormal model behaviors, such as repetitive outputs. To validate our approach, we conduct extensive evaluations, showing that Keye-VL achieves state-of-the-art results on public video benchmarks and remains highly competitive on general image-based tasks (Figure 1). Furthermore, we develop and release the \textbf{KC-MMBench}, a new benchmark tailored for real-world short-video scenarios, where Keye-VL shows a significant advantage.

Biao Yang, Bin Wen, Boyang Ding et al.

In recent years, the development of Large Language Models (LLMs) has significantly advanced, extending their capabilities to multimodal tasks through Multimodal Large Language Models (MLLMs). However, video understanding remains a challenging area due to the dynamic and information-dense nature of videos. Existing models struggle with the trade-off between spatial resolution and temporal coverage when processing video content. We present Keye-VL-1.5, which addresses fundamental challenges in video comprehension through three key innovations. First, we introduce a novel Slow-Fast video encoding strategy that dynamically allocates computational resources based on inter-frame similarity, processing key frames with significant visual changes at higher resolution (Slow pathway) while handling relatively static frames with increased temporal coverage at lower resolution (Fast pathway). Second, we implement a progressive four-stage pre-training methodology that systematically extends the model's context length from 8K to 128K tokens, enabling processing of longer videos and more complex visual content. Third, we develop a comprehensive post-training pipeline focusing on reasoning enhancement and human preference alignment, incorporating a 5-step chain-of-thought data construction process, iterative GSPO-based reinforcement learning with progressive prompt hinting for difficult cases, and alignment training. Through extensive evaluation on public benchmarks and rigorous internal human assessment, Keye-VL-1.5 demonstrates significant improvements over existing models, particularly excelling in video understanding tasks while maintaining competitive performance on general multimodal benchmarks.

Yingping Wang, Yi Wu, Xiangyu Wu et al.

Mixture-of-Experts (MoE) architectures are widely used in modern large language models and multimodal models. However, inference efficiency is often limited by highly dynamic and skewed expert workloads across different modalities. During the prefill stage with large batch sizes, vision tokens frequently dominate the input sequences. Under expert parallelism (EP), this leads to severe load imbalance, where a subset of devices becomes overloaded, reducing overall system throughput. We propose ReaLB, a real-time load balancing method for multimodal MoE (MMoE) inference that introduces zero scheduling overhead. ReaLB dynamically adjusts the computation precision of MoE experts at runtime on a per-EP-rank basis. For ranks dominated by vision-heavy experts, ReaLB assigns lower-precision computation to improve execution efficiency by exploiting FP4 Tensor Cores. ReaLB does not require redundant experts or additional memory allocation. Instead, it performs layer-wise expert precision transformation on the fly and hides the associated overhead within the dispatch phase before MoE computation. Experiments on representative MMoE models show that ReaLB achieves 1.29x layer-level speedup while limiting accuracy loss to within 1.2%.

Jiawei Chen, Ruoxi Xu, Boxi Cao et al.

The emergence of Large Language Models (LLMs) has illuminated the potential for a general-purpose user simulator. However, existing benchmarks remain constrained to isolated scenarios, narrow action spaces, or synthetic data, failing to capture the holistic nature of authentic human behavior. To bridge this gap, we introduce OmniBehavior, the first user simulation benchmark constructed entirely from real-world data, integrating long-horizon, cross-scenario, and heterogeneous behavioral patterns into a unified framework. Based on this benchmark, we first provide empirical evidence that previous datasets with isolated scenarios suffer from tunnel vision, whereas real-world decision-making relies on long-term, cross-scenario causal chains. Extensive evaluations of state-of-the-art LLMs reveal that current models struggle to accurately simulate these complex behaviors, with performance plateauing even as context windows expand. Crucially, a systematic comparison between simulated and authentic behaviors uncovers a fundamental structural bias: LLMs tend to converge toward a positive average person, exhibiting hyper-activity, persona homogenization, and a Utopian bias. This results in the loss of individual differences and long-tail behaviors, highlighting critical directions for future high-fidelity simulation research.

Dian Chao, Xin Song, Shupeng Zhong et al.

In this paper, we propose a solution for improving the quality of captions generated for figures in papers. We adopt the approach of summarizing the textual content in the paper to generate image captions. Throughout our study, we encounter discrepancies in the OCR information provided in the official dataset. To rectify this, we employ the PaddleOCR toolkit to extract OCR information from all images. Moreover, we observe that certain textual content in the official paper pertains to images that are not relevant for captioning, thereby introducing noise during caption generation. To mitigate this issue, we leverage LLaMA to extract image-specific information by querying the textual content based on image mentions, effectively filtering out extraneous information. Additionally, we recognize a discrepancy between the primary use of maximum likelihood estimation during text generation and the evaluation metrics such as ROUGE employed to assess the quality of generated captions. To bridge this gap, we integrate the BRIO model framework, enabling a more coherent alignment between the generation and evaluation processes. Our approach ranked first in the final test with a score of 4.49.

Can Xie, Ruotong Pan, Xiangyu Wu et al.

Reinforcement Learning with Verifiable Rewards (RLVR) has shown significant promise for enhancing the reasoning capabilities of large language models (LLMs). However, prevailing algorithms like GRPO broadcast a uniform advantage signal across all tokens in a sequence. This coarse-grained approach overlooks the pivotal role of uncertain, high-stakes decisions during reasoning, leading to inefficient exploration and the well-documented problem of entropy collapse. To address this, we introduce UnCertainty-aware Advantage Shaping (UCAS), a model-free method that refines credit assignment by leveraging the model's internal uncertainty signals. UCAS operates in two stages: it first modulates the response-level advantage using the model's overall self-confidence, and then applies a token-level penalty based on raw logit certainty. This dual mechanism encourages exploration of high-uncertainty paths that yield correct answers while penalizing overconfident yet erroneous reasoning, effectively balancing the exploration-exploitation trade-off. Extensive experiments on five mathematical reasoning benchmarks show that UCAS significantly outperforms strong RLVR baselines across multiple model scales, including 1.5B and 7B. Our analysis confirms that UCAS not only achieves higher rewards but also promotes greater reasoning diversity and successfully mitigates entropy collapse.

Longfei Huang, Shupeng Zhong, Xiangyu Wu et al.

This report introduces a solution to the Topic 1 Zero-shot Image Captioning of 2024 NICE : New frontiers for zero-shot Image Captioning Evaluation. In contrast to NICE 2023 datasets, this challenge involves new annotations by humans with significant differences in caption style and content. Therefore, we enhance image captions effectively through retrieval augmentation and caption grading methods. At the data level, we utilize high-quality captions generated by image caption models as training data to address the gap in text styles. At the model level, we employ OFA (a large-scale visual-language pre-training model based on handcrafted templates) to perform the image captioning task. Subsequently, we propose caption-level strategy for the high-quality caption data generated by the image caption models and integrate them with retrieval augmentation strategy into the template to compel the model to generate higher quality, more matching, and semantically enriched captions based on the retrieval augmentation prompts. Our approach achieves a CIDEr score of 234.11.

Marco Tranzatto, Frank Mascarich, Lukas Bernreiter et al.

Autonomous exploration of subterranean environments constitutes a major frontier for robotic systems as underground settings present key challenges that can render robot autonomy hard to achieve. This has motivated the DARPA Subterranean Challenge, where teams of robots search for objects of interest in various underground environments. In response, the CERBERUS system-of-systems is presented as a unified strategy towards subterranean exploration using legged and flying robots. As primary robots, ANYmal quadruped systems are deployed considering their endurance and potential to traverse challenging terrain. For aerial robots, both conventional and collision-tolerant multirotors are utilized to explore spaces too narrow or otherwise unreachable by ground systems. Anticipating degraded sensing conditions, a complementary multi-modal sensor fusion approach utilizing camera, LiDAR, and inertial data for resilient robot pose estimation is proposed. Individual robot pose estimates are refined by a centralized multi-robot map optimization approach to improve the reported location accuracy of detected objects of interest in the DARPA-defined coordinate frame. Furthermore, a unified exploration path planning policy is presented to facilitate the autonomous operation of both legged and aerial robots in complex underground networks. Finally, to enable communication between the robots and the base station, CERBERUS utilizes a ground rover with a high-gain antenna and an optical fiber connection to the base station, alongside breadcrumbing of wireless nodes by our legged robots. We report results from the CERBERUS system-of-systems deployment at the DARPA Subterranean Challenge Tunnel and Urban Circuits, along with the current limitations and the lessons learned for the benefit of the community.

Xing Cheng, Hezheng Lin, Xiangyu Wu et al.

Employing large-scale pre-trained model CLIP to conduct video-text retrieval task (VTR) has become a new trend, which exceeds previous VTR methods. Though, due to the heterogeneity of structures and contents between video and text, previous CLIP-based models are prone to overfitting in the training phase, resulting in relatively poor retrieval performance. In this paper, we propose a multi-stream Corpus Alignment network with single gate Mixture-of-Experts (CAMoE) and a novel Dual Softmax Loss (DSL) to solve the two heterogeneity. The CAMoE employs Mixture-of-Experts (MoE) to extract multi-perspective video representations, including action, entity, scene, etc., then align them with the corresponding part of the text. In this stage, we conduct massive explorations towards the feature extraction module and feature alignment module. DSL is proposed to avoid the one-way optimum-match which occurs in previous contrastive methods. Introducing the intrinsic prior of each pair in a batch, DSL serves as a reviser to correct the similarity matrix and achieves the dual optimal match. DSL is easy to implement with only one-line code but improves significantly. The results show that the proposed CAMoE and DSL are of strong efficiency, and each of them is capable of achieving State-of-The-Art (SOTA) individually on various benchmarks such as MSR-VTT, MSVD, and LSMDC. Further, with both of them, the performance is advanced to a big extend, surpassing the previous SOTA methods for around 4.6\% R@1 in MSR-VTT.

Xiangyu Wu, Jun Zeng, Andrea Tagliabue et al.

Limited flight distance and time is a common problem for multicopters. We propose a method for finding the optimal speed and sideslip angle of a multicopter flying a given path to achieve either the longest flight distance or time. Since flight speed and sideslip are often free variables in multicopter path planning, they can be changed without changing the mission. The proposed method is based on a novel multivariable extremum seeking controller with adaptive step size, which is inspired by recent work from the machine learning community on stochastic optimization. Our method (a) does not require a power consumption model of the vehicle, (b) is computationally efficient and runs on low-cost embedded computers in real-time, and (c) converges faster than the standard extremum seeking controller with constant step size. We prove the stability of this approach and validate it through outdoor experiments. The method is shown to converge with different payloads and in the presence of wind. Compared to flying at the maximum achievable speed in the experiments with a uniformly selected random sideslip angle, flying at the optimal range speed and sideslip on average increases the flight range by 14.3% without payload and 19.4% with a box payload. In addition, compared to hovering, flying at the optimal endurance speed and sideslip increases the flight time by 7.5% without payload and 14.4% with a box payload. A video can be found at https://youtu.be/aLds8LVfogk.

Shuxiao Chen, Xiangyu Wu, Mark W. Mueller et al.

The capabilities of autonomous flight with unmanned aerial vehicles (UAVs) have significantly increased in recent times. However, basic problems such as fast and robust geo-localization in GPS-denied environments still remain unsolved. Existing research has primarily concentrated on improving the accuracy of localization at the cost of long and varying computation time in various situations, which often necessitates the use of powerful ground station machines. In order to make image-based geo-localization online and pragmatic for lightweight embedded systems on UAVs, we propose a framework that is reliable in changing scenes, flexible about computing resource allocation and adaptable to common camera placements. The framework is comprised of two stages: offline database preparation and online inference. At the first stage, color images and depth maps are rendered as seen from potential vehicle poses quantized over the satellite and topography maps of anticipated flying areas. A database is then populated with the global and local descriptors of the rendered images. At the second stage, for each captured real-world query image, top global matches are retrieved from the database and the vehicle pose is further refined via local descriptor matching. We present field experiments of image-based localization on two different UAV platforms to validate our results.

Junseok Lee, Xiangyu Wu, Seung Jae Lee et al.

This paper introduces a collision avoidance system for navigating a multicopter in cluttered outdoor environments based on the recent memory-less motion planner, rectangular pyramid partitioning using integrated depth sensors (RAPPIDS). The RAPPIDS motion planner generates collision-free flight trajectories at high speed with low computational cost using only the latest depth image. In this work we extend it to improve the performance of the planner by taking the following issues into account. (a) Changes in the dynamic characteristics of the multicopter that occur during flight, such as changes in motor input/output characteristics due to battery voltage drop. (b) The noise of the flight sensor, which can cause unwanted control input components. (c) Planner utility function which may not be suitable for the cluttered environment. Therefore, in this paper we introduce solutions to each of the above problems and propose a system for the successful operation of the RAPPIDS planner in an outdoor cluttered flight environment. At the end of the paper, we validate the proposed method's effectiveness by presenting the flight experiment results in a forest environment. A video can be found at www.youtube.com/channel/UCK-gErmvZlBODN5gQpNcpsg

Jiaming Zha, Xiangyu Wu, Joseph Kroeger et al.

Aerial vehicles with collision resilience can operate with more confidence in environments with obstacles that are hard to detect and avoid. This paper presents the methodology used to design a collision resilient aerial vehicle with icosahedron tensegrity structure. A simplified stress analysis of the tensegrity frame under impact forces is performed to guide the selection of its components. In addition, an autonomous controller is presented to reorient the vehicle from an arbitrary orientation on the ground to help it take off. Experiments show that the vehicle can successfully reorient itself after landing upside-down and can survive collisions with speed up to 6.5m/s.

Xiangyu Wu, Mark W. Mueller

Limited flight range is a common problem for multicopters. To alleviate this problem, we propose a method for finding the optimal speed and heading of a multicopter when flying a given path to achieve the longest flight range. Based on a novel multivariable extremum seeking controller with adaptive step size, the method (a) does not require any power consumption model of the vehicle, (b) can adapt to unknown disturbances, (c) can be executed online, and (d) converges faster than the standard extremum seeking controller with constant step size. We conducted indoor experiments to validate the effectiveness of this method under different payloads and initial conditions, and showed that it is able to converge more than 30% faster than the standard extremum seeking controller. This method is especially useful for applications such as package delivery, where the size and weight of the payload differ for different deliveries and the power consumption of the vehicle is hard to model.