Bin Li

Yixuan Weng, Minjun Zhu, Fei Xia et al.

Recently, with the chain of thought (CoT) prompting, large language models (LLMs), e.g., GPT-3, have shown strong reasoning ability in several natural language processing tasks such as arithmetic, commonsense, and logical reasoning. However, LLMs with CoT require multi-step prompting and multi-token prediction, which is highly sensitive to individual mistakes and vulnerable to error accumulation. The above issues make the LLMs need the ability to verify the answers. In fact, after inferring conclusions in some thinking decision tasks, people often check them by re-verifying steps to avoid some mistakes. In this paper, we propose and prove that LLMs also have similar self-verification abilities. We take the conclusion obtained by CoT as one of the conditions for solving the original problem. By performing a backward verification of the answers that LLM deduced for itself, we can obtain interpretable answer validation scores to select the candidate answer with the highest score. Experimental results demonstrate that the proposed method can improve the reasoning performance on various arithmetic, commonsense, and logical reasoning datasets. Our code is publicly available at: https://github.com/WENGSYX/Self-Verification.

Jiahao Li, Wenxuan Xie, Zhaoyang Jia et al.

While neural video codecs (NVCs) have demonstrated superior compression ratio, their prohibitive computational complexity remains a critical barrier to real-world deployment. This paper introduces a chunk-based coding framework designed to significantly improve the rate-distortion-complexity trade-off. Instead of processing frames sequentially, our approach encodes a chunk of multiple frames into a single compact latent representation and decodes them simultaneously. This is enabled by cross-frame interaction modules for joint spatial-temporal modeling and frame-specific decoders for parallel reconstruction. This paradigm not only dramatically enhances coding throughput but also facilitates more effective modeling of long-term temporal correlations. To further boost speed, we propose a streamlined entropy coding mechanism that consolidates bit-stream interactions into a single step, substantially reducing decoding overhead. Building on these innovations, we present DCVC-UF (Ultra-Fast), a new NVC that sets a new SOTA in performance. Our experiments show that DCVC-UF can achieve ultra-fast encoding and decoding speeds, significantly outperforming previous leading codecs. DCVC-UF serves as a notable landmark in the journey of NVC evolution. The code is at https://github.com/microsoft/DCVC.

Chenxin Tao, Xizhou Zhu, Weijie Su et al.

Self-supervised learning (SSL) has delivered superior performance on a variety of downstream vision tasks. Two main-stream SSL frameworks have been proposed, i.e., Instance Discrimination (ID) and Masked Image Modeling (MIM). ID pulls together representations from different views of the same image, while avoiding feature collapse. It lacks spatial sensitivity, which requires modeling the local structure within each image. On the other hand, MIM reconstructs the original content given a masked image. It instead does not have good semantic alignment, which requires projecting semantically similar views into nearby representations. To address this dilemma, we observe that (1) semantic alignment can be achieved by matching different image views with strong augmentations; (2) spatial sensitivity can benefit from predicting dense representations with masked images. Driven by these analysis, we propose Siamese Image Modeling (SiameseIM), which predicts the dense representations of an augmented view, based on another masked view from the same image but with different augmentations. SiameseIM uses a Siamese network with two branches. The online branch encodes the first view, and predicts the second view's representation according to the relative positions between these two views. The target branch produces the target by encoding the second view. SiameseIM can surpass both ID and MIM on a wide range of downstream tasks, including ImageNet finetuning and linear probing, COCO and LVIS detection, and ADE20k semantic segmentation. The improvement is more significant in few-shot, long-tail and robustness-concerned scenarios. Code shall be released at https://github.com/fundamentalvision/Siamese-Image-Modeling.

Weijie Su, Xizhou Zhu, Chenxin Tao et al.

To effectively exploit the potential of large-scale models, various pre-training strategies supported by massive data from different sources are proposed, including supervised pre-training, weakly-supervised pre-training, and self-supervised pre-training. It has been proved that combining multiple pre-training strategies and data from various modalities/sources can greatly boost the training of large-scale models. However, current works adopt a multi-stage pre-training system, where the complex pipeline may increase the uncertainty and instability of the pre-training. It is thus desirable that these strategies can be integrated in a single-stage manner. In this paper, we first propose a general multi-modal mutual information formula as a unified optimization target and demonstrate that all existing approaches are special cases of our framework. Under this unified perspective, we propose an all-in-one single-stage pre-training approach, named Maximizing Multi-modal Mutual Information Pre-training (M3I Pre-training). Our approach achieves better performance than previous pre-training methods on various vision benchmarks, including ImageNet classification, COCO object detection, LVIS long-tailed object detection, and ADE20k semantic segmentation. Notably, we successfully pre-train a billion-level parameter image backbone and achieve state-of-the-art performance on various benchmarks. Code shall be released at https://github.com/OpenGVLab/M3I-Pretraining.

Pengfei Xia, Ziqiang Li, Wei Zhang et al.

Recent studies have proven that deep neural networks are vulnerable to backdoor attacks. Specifically, by mixing a small number of poisoned samples into the training set, the behavior of the trained model can be maliciously controlled. Existing attack methods construct such adversaries by randomly selecting some clean data from the benign set and then embedding a trigger into them. However, this selection strategy ignores the fact that each poisoned sample contributes inequally to the backdoor injection, which reduces the efficiency of poisoning. In this paper, we formulate improving the poisoned data efficiency by the selection as an optimization problem and propose a Filtering-and-Updating Strategy (FUS) to solve it. The experimental results on CIFAR-10 and ImageNet-10 indicate that the proposed method is effective: the same attack success rate can be achieved with only 47% to 75% of the poisoned sample volume compared to the random selection strategy. More importantly, the adversaries selected according to one setting can generalize well to other settings, exhibiting strong transferability. The prototype code of our method is now available at https://github.com/xpf/Data-Efficient-Backdoor-Attacks.

Shangchao Su, Bin Li, Xiangyang Xue

Federated Learning (FL) has recently made significant progress as a new machine learning paradigm for privacy protection. Due to the high communication cost of traditional FL, one-shot federated learning is gaining popularity as a way to reduce communication cost between clients and the server. Most of the existing one-shot FL methods are based on Knowledge Distillation; however, {distillation based approach requires an extra training phase and depends on publicly available data sets or generated pseudo samples.} In this work, we consider a novel and challenging cross-silo setting: performing a single round of parameter aggregation on the local models without server-side training. In this setting, we propose an effective algorithm for Model Aggregation via Exploring Common Harmonized Optima (MA-Echo), which iteratively updates the parameters of all local models to bring them close to a common low-loss area on the loss surface, without harming performance on their own data sets at the same time. Compared to the existing methods, MA-Echo can work well even in extremely non-identical data distribution settings where the support categories of each local model have no overlapped labels with those of the others. We conduct extensive experiments on two popular image classification data sets to compare the proposed method with existing methods and demonstrate the effectiveness of MA-Echo, which clearly outperforms the state-of-the-arts. The source code can be accessed in \url{https://github.com/FudanVI/MAEcho}.

Jiahao Li, Bin Li, Yan Lu

For neural video codec, it is critical, yet challenging, to design an efficient entropy model which can accurately predict the probability distribution of the quantized latent representation. However, most existing video codecs directly use the ready-made entropy model from image codec to encode the residual or motion, and do not fully leverage the spatial-temporal characteristics in video. To this end, this paper proposes a powerful entropy model which efficiently captures both spatial and temporal dependencies. In particular, we introduce the latent prior which exploits the correlation among the latent representation to squeeze the temporal redundancy. Meanwhile, the dual spatial prior is proposed to reduce the spatial redundancy in a parallel-friendly manner. In addition, our entropy model is also versatile. Besides estimating the probability distribution, our entropy model also generates the quantization step at spatial-channel-wise. This content-adaptive quantization mechanism not only helps our codec achieve the smooth rate adjustment in single model but also improves the final rate-distortion performance by dynamic bit allocation. Experimental results show that, powered by the proposed entropy model, our neural codec can achieve 18.2% bitrate saving on UVG dataset when compared with H.266 (VTM) using the highest compression ratio configuration. It makes a new milestone in the development of neural video codec. The codes are at https://github.com/microsoft/DCVC.

Fei Xia, Bin Li, Yixuan Weng et al.

The medical conversational system can relieve the burden of doctors and improve the efficiency of healthcare, especially during the pandemic. This paper presents a medical conversational question answering (CQA) system based on the multi-modal knowledge graph, namely "LingYi", which is designed as a pipeline framework to maintain high flexibility. Our system utilizes automated medical procedures including medical triage, consultation, image-text drug recommendation and record. To conduct knowledge-grounded dialogues with patients, we first construct a Chinese Medical Multi-Modal Knowledge Graph (CM3KG) and collect a large-scale Chinese Medical CQA (CMCQA) dataset. Compared with the other existing medical question-answering systems, our system adopts several state-of-the-art technologies including medical entity disambiguation and medical dialogue generation, which is more friendly to provide medical services to patients. In addition, we have open-sourced our codes which contain back-end models and front-end web pages at https://github.com/WENGSYX/LingYi. The datasets including CM3KG at https://github.com/WENGSYX/CM3KG and CMCQA at https://github.com/WENGSYX/CMCQA are also released to further promote future research.

Guo-Hua Wang, Jiahao Li, Bin Li et al.

Neural image compression has surpassed state-of-the-art traditional codecs (H.266/VVC) for rate-distortion (RD) performance, but suffers from large complexity and separate models for different rate-distortion trade-offs. In this paper, we propose an Efficient single-model Variable-bit-rate Codec (EVC), which is able to run at 30 FPS with 768x512 input images and still outperforms VVC for the RD performance. By further reducing both encoder and decoder complexities, our small model even achieves 30 FPS with 1920x1080 input images. To bridge the performance gap between our different capacities models, we meticulously design the mask decay, which transforms the large model's parameters into the small model automatically. And a novel sparsity regularization loss is proposed to mitigate shortcomings of $L_p$ regularization. Our algorithm significantly narrows the performance gap by 50% and 30% for our medium and small models, respectively. At last, we advocate the scalable encoder for neural image compression. The encoding complexity is dynamic to meet different latency requirements. We propose decaying the large encoder multiple times to reduce the residual representation progressively. Both mask decay and residual representation learning greatly improve the RD performance of our scalable encoder. Our code is at https://github.com/microsoft/DCVC.

Kang Chen, Tao Han, Junchao Gong et al.

We present FengWu, an advanced data-driven global medium-range weather forecast system based on Artificial Intelligence (AI). Different from existing data-driven weather forecast methods, FengWu solves the medium-range forecast problem from a multi-modal and multi-task perspective. Specifically, a deep learning architecture equipped with model-specific encoder-decoders and cross-modal fusion Transformer is elaborately designed, which is learned under the supervision of an uncertainty loss to balance the optimization of different predictors in a region-adaptive manner. Besides this, a replay buffer mechanism is introduced to improve medium-range forecast performance. With 39-year data training based on the ERA5 reanalysis, FengWu is able to accurately reproduce the atmospheric dynamics and predict the future land and atmosphere states at 37 vertical levels on a 0.25° latitude-longitude resolution. Hindcasts of 6-hourly weather in 2018 based on ERA5 demonstrate that FengWu performs better than GraphCast in predicting 80\% of the 880 reported predictands, e.g., reducing the root mean square error (RMSE) of 10-day lead global z500 prediction from 733 to 651 $m^{2}/s^2$. In addition, the inference cost of each iteration is merely 600ms on NVIDIA Tesla A100 hardware. The results suggest that FengWu can significantly improve the forecast skill and extend the skillful global medium-range weather forecast out to 10.75 days lead (with ACC of z500 > 0.6) for the first time.

Jiahao Li, Bin Li, Yan Lu

For any video codecs, the coding efficiency highly relies on whether the current signal to be encoded can find the relevant contexts from the previous reconstructed signals. Traditional codec has verified more contexts bring substantial coding gain, but in a time-consuming manner. However, for the emerging neural video codec (NVC), its contexts are still limited, leading to low compression ratio. To boost NVC, this paper proposes increasing the context diversity in both temporal and spatial dimensions. First, we guide the model to learn hierarchical quality patterns across frames, which enriches long-term and yet high-quality temporal contexts. Furthermore, to tap the potential of optical flow-based coding framework, we introduce a group-based offset diversity where the cross-group interaction is proposed for better context mining. In addition, this paper also adopts a quadtree-based partition to increase spatial context diversity when encoding the latent representation in parallel. Experiments show that our codec obtains 23.5% bitrate saving over previous SOTA NVC. Better yet, our codec has surpassed the under-developing next generation traditional codec/ECM in both RGB and YUV420 colorspaces, in terms of PSNR. The codes are at https://github.com/microsoft/DCVC.

Jinghan Zhao, Wenwei Jin, Anqi Li et al.

Item-to-Item (I2I) retrieval is a fundamental part of modern content platforms, supporting critical industrial workflows from recommendation engines to content auditing. While multimodal embedding methods have advanced general retrieval, they often falter in I2I scenarios due to the challenges of balancing global content representation with fine-grained local retrieval, the systemic inefficiency of decoupled embedding-and-ranking pipelines, and the inherent trade-offs between model precision and serving latency. To solve these issues, we propose \textbf{UniNote}, a unified embedding model designed for industrial I2I retrieval. Tailored retrieval strategies are introduced to support representation learning over complex, multimodal content at varying granularities. To operationalize these strategies, UniNote employs a two-stage training paradigm: the first stage leverages contrastive SFT to establish robust base embeddings, while the second stage refines ranking quality through a reinforcement learning (RL) process that aligns the model with content relevance. Our results show that UniNote achieves SOTA performance across diverse I2I tasks. Deployed at Xiaohongshu and integrated with Matryoshka Representation Learning (MRL), UniNote achieved significant improvements in retrieval quality and cost efficiency in large-scale applications.

Jiafeng Zhong, Bin Li, Jiangyan Yi

The task of partially spoofed audio localization aims to accurately determine audio authenticity at a frame level. Although some works have achieved encouraging results, utilizing boundary information within a single model remains an unexplored research topic. In this work, we propose a novel method called Boundary-aware Attention Mechanism (BAM). Specifically, it consists of two core modules: Boundary Enhancement and Boundary Frame-wise Attention. The former assembles the intra-frame and inter-frame information to extract discriminative boundary features that are subsequently used for boundary position detection and authenticity decision, while the latter leverages boundary prediction results to explicitly control the feature interaction between frames, which achieves effective discrimination between real and fake frames. Experimental results on PartialSpoof database demonstrate our proposed method achieves the best performance. The code is available at https://github.com/media-sec-lab/BAM.

Shangchao Su, Mingzhao Yang, Bin Li et al.

Federated learning (FL) enables multiple clients to collaboratively train a global model without disclosing their data. Previous researches often require training the complete model parameters. However, the emergence of powerful pre-trained models makes it possible to achieve higher performance with fewer learnable parameters in FL. In this paper, we propose a federated adaptive prompt tuning algorithm, FedAPT, for multi-domain collaborative image classification with powerful foundation models, like CLIP. Compared with direct federated prompt tuning, our core idea is to adaptively unlock specific domain knowledge for each test sample in order to provide them with personalized prompts. To implement this idea, we design an adaptive prompt tuning module, which consists of a meta prompt, an adaptive network, and some keys. The server randomly generates a set of keys and assigns a unique key to each client. Then all clients cooperatively train the global adaptive network and meta prompt with the local datasets and the frozen keys. Ultimately, the global aggregation model can assign a personalized prompt to CLIP based on the domain features of each test sample. We perform extensive experiments on two multi-domain image classification datasets across two different settings -- supervised and unsupervised. The results show that FedAPT can achieve better performance with less than 10\% of the number of parameters of the fully trained model, and the global model can perform well in diverse client domains simultaneously. The source code is available at \url{https://github.com/leondada/FedAPT}.

Yu Zhu, Yongkang Li, Wenjie Zhu et al.

Recent studies have explored Vision-Language Models (VLMs) for food analysis. However, most existing methods rely primarily on supervised fine-tuning (SFT), which often limits reasoning and generalization capabilities. Moreover, high-quality large-scale nutritional annotations remain scarce. To address these issues, we introduce CalorieBench-80K, a large-scale benchmark with curated calorie labels and dietary advice annotations. To the best of our knowledge, it is the first food image benchmark to incorporate Chain-of-Thought (CoT) annotations for calorie reasoning. We also propose Food-R1, a unified food VLM trained in a multi-task learning paradigm to equip the model with broad capabilities. Food-R1 undergoes CoT-based cold-start instruction tuning, followed by reinforcement fine-tuning (RFT) using Group Relative Policy Optimization (GRPO) to improve reasoning and performance. Experiments on CalorieBench-80K and representative benchmarks show that Food-R1 consistently outperforms strong baselines across food-related tasks. The code, model weights, and benchmark annotations are available at the project repository.

Tao Huang, Kai Chen, Bin Li et al.

Task automation of surgical robot has the potentials to improve surgical efficiency. Recent reinforcement learning (RL) based approaches provide scalable solutions to surgical automation, but typically require extensive data collection to solve a task if no prior knowledge is given. This issue is known as the exploration challenge, which can be alleviated by providing expert demonstrations to an RL agent. Yet, how to make effective use of demonstration data to improve exploration efficiency still remains an open challenge. In this work, we introduce Demonstration-guided EXploration (DEX), an efficient reinforcement learning algorithm that aims to overcome the exploration problem with expert demonstrations for surgical automation. To effectively exploit demonstrations, our method estimates expert-like behaviors with higher values to facilitate productive interactions, and adopts non-parametric regression to enable such guidance at states unobserved in demonstration data. Extensive experiments on $10$ surgical manipulation tasks from SurRoL, a comprehensive surgical simulation platform, demonstrate significant improvements in the exploration efficiency and task success rates of our method. Moreover, we also deploy the learned policies to the da Vinci Research Kit (dVRK) platform to show the effectiveness on the real robot. Code is available at https://github.com/med-air/DEX.

Qizao Wang, Xuelin Qian, Bin Li et al.

Cloth-changing person Re-IDentification (Re-ID) is a particularly challenging task, suffering from two limitations of inferior discriminative features and limited training samples. Existing methods mainly leverage auxiliary information to facilitate identity-relevant feature learning, including soft-biometrics features of shapes or gaits, and additional labels of clothing. However, this information may be unavailable in real-world applications. In this paper, we propose a novel FIne-grained Representation and Recomposition (FIRe$^{2}$) framework to tackle both limitations without any auxiliary annotation or data. Specifically, we first design a Fine-grained Feature Mining (FFM) module to separately cluster images of each person. Images with similar so-called fine-grained attributes (e.g., clothes and viewpoints) are encouraged to cluster together. An attribute-aware classification loss is introduced to perform fine-grained learning based on cluster labels, which are not shared among different people, promoting the model to learn identity-relevant features. Furthermore, to take full advantage of fine-grained attributes, we present a Fine-grained Attribute Recomposition (FAR) module by recomposing image features with different attributes in the latent space. It significantly enhances robust feature learning. Extensive experiments demonstrate that FIRe$^{2}$ can achieve state-of-the-art performance on five widely-used cloth-changing person Re-ID benchmarks. The code is available at https://github.com/QizaoWang/FIRe-CCReID.

Hongjing Niu, Hanting Li, Feng Zhao et al.

Domain generalization (DG), aiming to make models work on unseen domains, is a surefire way toward general artificial intelligence. Limited by the scale and diversity of current DG datasets, it is difficult for existing methods to scale to diverse domains in open-world scenarios (e.g., science fiction and pixelate style). Therefore, the source-free domain generalization (SFDG) task is necessary and challenging. To address this issue, we propose an approach based on large-scale vision-language pretraining models (e.g., CLIP), which exploits the extensive domain information embedded in it. The proposed scheme generates diverse prompts from a domain bank that contains many more diverse domains than existing DG datasets. Furthermore, our method yields domain-unified representations from these prompts, thus being able to cope with samples from open-world domains. Extensive experiments on mainstream DG datasets, namely PACS, VLCS, OfficeHome, and DomainNet, show that the proposed method achieves competitive performance compared to state-of-the-art (SOTA) DG methods that require source domain data for training. Besides, we collect a small datasets consists of two domains to evaluate the open-world domain generalization ability of the proposed method. The source code and the dataset will be made publicly available at https://github.com/muse1998/Source-Free-Domain-Generalization

Ziqiang Li, Chaoyue Wang, Heliang Zheng et al.

Data-Efficient GANs (DE-GANs), which aim to learn generative models with a limited amount of training data, encounter several challenges for generating high-quality samples. Since data augmentation strategies have largely alleviated the training instability, how to further improve the generative performance of DE-GANs becomes a hotspot. Recently, contrastive learning has shown the great potential of increasing the synthesis quality of DE-GANs, yet related principles are not well explored. In this paper, we revisit and compare different contrastive learning strategies in DE-GANs, and identify (i) the current bottleneck of generative performance is the discontinuity of latent space; (ii) compared to other contrastive learning strategies, Instance-perturbation works towards latent space continuity, which brings the major improvement to DE-GANs. Based on these observations, we propose FakeCLR, which only applies contrastive learning on perturbed fake samples, and devises three related training techniques: Noise-related Latent Augmentation, Diversity-aware Queue, and Forgetting Factor of Queue. Our experimental results manifest the new state of the arts on both few-shot generation and limited-data generation. On multiple datasets, FakeCLR acquires more than 15% FID improvement compared to existing DE-GANs. Code is available at https://github.com/iceli1007/FakeCLR.

Zhaoyang Jia, Jiahao Li, Bin Li et al.

Most existing image compression approaches perform transform coding in the pixel space to reduce its spatial redundancy. However, they encounter difficulties in achieving both high-realism and high-fidelity at low bitrate, as the pixel-space distortion may not align with human perception. To address this issue, we introduce a Generative Latent Coding (GLC) architecture, which performs transform coding in the latent space of a generative vector-quantized variational auto-encoder (VQ-VAE), instead of in the pixel space. The generative latent space is characterized by greater sparsity, richer semantic and better alignment with human perception, rendering it advantageous for achieving high-realism and high-fidelity compression. Additionally, we introduce a categorical hyper module to reduce the bit cost of hyper-information, and a code-prediction-based supervision to enhance the semantic consistency. Experiments demonstrate that our GLC maintains high visual quality with less than 0.04 bpp on natural images and less than 0.01 bpp on facial images. On the CLIC2020 test set, we achieve the same FID as MS-ILLM with 45% fewer bits. Furthermore, the powerful generative latent space enables various applications built on our GLC pipeline, such as image restoration and style transfer. The code is available at https://github.com/jzyustc/GLC.

Yi Wu, Ziqiang Li, Chaoyue Wang et al.

In this study, we delve into the task of few-shot Generative Domain Adaptation (GDA), which involves transferring a pre-trained generator from one domain to a new domain using only a few reference images. Inspired by the way human brains acquire knowledge in new domains, we present an innovative generator structure called Domain Re-Modulation (DoRM). DoRM not only meets the criteria of high quality, large synthesis diversity, and cross-domain consistency, which were achieved by previous research in GDA, but also incorporates memory and domain association, akin to how human brains operate. Specifically, DoRM freezes the source generator and introduces new mapping and affine modules (M&A modules) to capture the attributes of the target domain during GDA. This process resembles the formation of new synapses in human brains. Consequently, a linearly combinable domain shift occurs in the style space. By incorporating multiple new M&A modules, the generator gains the capability to perform high-fidelity multi-domain and hybrid-domain generation. Moreover, to maintain cross-domain consistency more effectively, we introduce a similarity-based structure loss. This loss aligns the auto-correlation map of the target image with its corresponding auto-correlation map of the source image during training. Through extensive experiments, we demonstrate the superior performance of our DoRM and similarity-based structure loss in few-shot GDA, both quantitatively and qualitatively. The code will be available at https://github.com/wuyi2020/DoRM.

Kang Chen, Shaochen Wang, Beihao Xia et al.

Transparent objects are widely used in industrial automation and daily life. However, robust visual recognition and perception of transparent objects have always been a major challenge. Currently, most commercial-grade depth cameras are still not good at sensing the surfaces of transparent objects due to the refraction and reflection of light. In this work, we present a transformer-based transparent object depth estimation approach from a single RGB-D input. We observe that the global characteristics of the transformer make it easier to extract contextual information to perform depth estimation of transparent areas. In addition, to better enhance the fine-grained features, a feature fusion module (FFM) is designed to assist coherent prediction. Our empirical evidence demonstrates that our model delivers significant improvements in recent popular datasets, e.g., 25% gain on RMSE and 21% gain on REL compared to previous state-of-the-art convolutional-based counterparts in ClearGrasp dataset. Extensive results show that our transformer-based model enables better aggregation of the object's RGB and inaccurate depth information to obtain a better depth representation. Our code and the pre-trained model will be available at https://github.com/yuchendoudou/TODE.

Bin Li, Yixuan Weng, Fei Xia et al.

The last decade has witnessed enormous improvements in science and technology, stimulating the growing demand for economic and cultural exchanges in various countries. Building a neural machine translation (NMT) system has become an urgent trend, especially in the low-resource setting. However, recent work tends to study NMT systems for low-resource languages centered on English, while few works focus on low-resource NMT systems centered on other languages such as Chinese. To achieve this, the low-resource multilingual translation challenge of the 2021 iFLYTEK AI Developer Competition provides the Chinese-centric multilingual low-resource NMT tasks, where participants are required to build NMT systems based on the provided low-resource samples. In this paper, we present the winner competition system that leverages monolingual word embeddings data enhancement, bilingual curriculum learning, and contrastive re-ranking. In addition, a new Incomplete-Trust (In-trust) loss function is proposed to replace the traditional cross-entropy loss when training. The experimental results demonstrate that the implementation of these ideas leads better performance than other state-of-the-art methods. All the experimental codes are released at: https://github.com/WENGSYX/Low-resource-text-translation.

Dongshuo Yin, Leiyi Hu, Bin Li et al.

Pre-training & fine-tuning can enhance the transferring efficiency and performance in visual tasks. Recent delta-tuning methods provide more options for visual classification tasks. Despite their success, existing visual delta-tuning art fails to exceed the upper limit of full fine-tuning on challenging tasks like object detection and segmentation. To find a competitive alternative to full fine-tuning, we propose the Multi-cognitive Visual Adapter (Mona) tuning, a novel adapter-based tuning method. First, we introduce multiple vision-friendly filters into the adapter to enhance its ability to process visual signals, while previous methods mainly rely on language-friendly linear filters. Second, we add the scaled normalization layer in the adapter to regulate the distribution of input features for visual filters. To fully demonstrate the practicality and generality of Mona, we conduct experiments on multiple representative visual tasks, including instance segmentation on COCO, semantic segmentation on ADE20K, object detection on Pascal VOC, oriented object detection on DOTA/STAR, and image classification on three common datasets. Exciting results illustrate that Mona surpasses full fine-tuning on all these tasks, and is the only delta-tuning method outperforming full fine-tuning on the above various tasks. For example, Mona achieves 1% performance gain on the COCO dataset compared to full fine-tuning. Comprehensive results suggest that Mona-tuning is more suitable for retaining and utilizing the capabilities of pre-trained models than full fine-tuning. The code will be released at https://github.com/Leiyi-Hu/mona.

Shangchao Su, Bin Li, Xiangyang Xue

With the increasing availability of Foundation Models, federated tuning has garnered attention in the field of federated learning, utilizing data and computation resources from multiple clients to collaboratively fine-tune foundation models. However, in real-world federated scenarios, there often exist a multitude of heterogeneous clients with varying computation and communication resources, rendering them incapable of supporting the entire model fine-tuning process. In response to this challenge, we propose a novel federated tuning algorithm, FedRA. The implementation of FedRA is straightforward and can be seamlessly integrated into any transformer-based model without the need for further modification to the original model. Specifically, in each communication round, FedRA randomly generates an allocation matrix. For resource-constrained clients, it reorganizes a small number of layers from the original model based on the allocation matrix and fine-tunes using adapters. Subsequently, the server aggregates the updated adapter parameters from the clients according to the current allocation matrix into the corresponding layers of the original model. It is worth noting that FedRA also supports scenarios where none of the clients can support the entire global model, which is an impressive advantage. We conduct experiments on two large-scale image datasets, DomainNet and NICO++, under various non-iid settings. The results demonstrate that FedRA outperforms the compared methods significantly. The source code is available at \url{https://github.com/leondada/FedRA}.

Ziqiang Li, Hong Sun, Pengfei Xia et al.

Recent deep neural networks (DNNs) have came to rely on vast amounts of training data, providing an opportunity for malicious attackers to exploit and contaminate the data to carry out backdoor attacks. However, existing backdoor attack methods make unrealistic assumptions, assuming that all training data comes from a single source and that attackers have full access to the training data. In this paper, we introduce a more realistic attack scenario where victims collect data from multiple sources, and attackers cannot access the complete training data. We refer to this scenario as data-constrained backdoor attacks. In such cases, previous attack methods suffer from severe efficiency degradation due to the entanglement between benign and poisoning features during the backdoor injection process. To tackle this problem, we introduce three CLIP-based technologies from two distinct streams: Clean Feature Suppression and Poisoning Feature Augmentation.effective solution for data-constrained backdoor attacks. The results demonstrate remarkable improvements, with some settings achieving over 100% improvement compared to existing attacks in data-constrained scenarios. Code is available at https://github.com/sunh1113/Efficient-backdoor-attacks-for-deep-neural-networks-in-real-world-scenarios

Rui Yang, Jie Wang, Zijie Geng et al.

Generalization across different environments with the same tasks is critical for successful applications of visual reinforcement learning (RL) in real scenarios. However, visual distractions -- which are common in real scenes -- from high-dimensional observations can be hurtful to the learned representations in visual RL, thus degrading the performance of generalization. To tackle this problem, we propose a novel approach, namely Characteristic Reward Sequence Prediction (CRESP), to extract the task-relevant information by learning reward sequence distributions (RSDs), as the reward signals are task-relevant in RL and invariant to visual distractions. Specifically, to effectively capture the task-relevant information via RSDs, CRESP introduces an auxiliary task -- that is, predicting the characteristic functions of RSDs -- to learn task-relevant representations, because we can well approximate the high-dimensional distributions by leveraging the corresponding characteristic functions. Experiments demonstrate that CRESP significantly improves the performance of generalization on unseen environments, outperforming several state-of-the-arts on DeepMind Control tasks with different visual distractions.

Teng Fu, Xiaocong Wang, Haiyang Yu et al.

Multiple object tracking (MOT) tends to become more challenging when severe occlusions occur. In this paper, we analyze the limitations of traditional Convolutional Neural Network-based methods and Transformer-based methods in handling occlusions and propose DNMOT, an end-to-end trainable DeNoising Transformer for MOT. To address the challenge of occlusions, we explicitly simulate the scenarios when occlusions occur. Specifically, we augment the trajectory with noises during training and make our model learn the denoising process in an encoder-decoder architecture, so that our model can exhibit strong robustness and perform well under crowded scenes. Additionally, we propose a Cascaded Mask strategy to better coordinate the interaction between different types of queries in the decoder to prevent the mutual suppression between neighboring trajectories under crowded scenes. Notably, the proposed method requires no additional modules like matching strategy and motion state estimation in inference. We conduct extensive experiments on the MOT17, MOT20, and DanceTrack datasets, and the experimental results show that our method outperforms previous state-of-the-art methods by a clear margin.

Tao Yan, Mingyue Li, Bin Li et al.

Existing deraining methods focus mainly on a single input image. However, with just a single input image, it is extremely difficult to accurately detect and remove rain streaks, in order to restore a rain-free image. In contrast, a light field image (LFI) embeds abundant 3D structure and texture information of the target scene by recording the direction and position of each incident ray via a plenoptic camera. LFIs are becoming popular in the computer vision and graphics communities. However, making full use of the abundant information available from LFIs, such as 2D array of sub-views and the disparity map of each sub-view, for effective rain removal is still a challenging problem. In this paper, we propose a novel method, 4D-MGP-SRRNet, for rain streak removal from LFIs. Our method takes as input all sub-views of a rainy LFI. To make full use of the LFI, it adopts 4D convolutional layers to simultaneously process all sub-views of the LFI. In the pipeline, the rain detection network, MGPDNet, with a novel Multi-scale Self-guided Gaussian Process (MSGP) module is proposed to detect high-resolution rain streaks from all sub-views of the input LFI at multi-scales. Semi-supervised learning is introduced for MSGP to accurately detect rain streaks by training on both virtual-world rainy LFIs and real-world rainy LFIs at multi-scales via computing pseudo ground truths for real-world rain streaks. We then feed all sub-views subtracting the predicted rain streaks into a 4D convolution-based Depth Estimation Residual Network (DERNet) to estimate the depth maps, which are later converted into fog maps. Finally, all sub-views concatenated with the corresponding rain streaks and fog maps are fed into a powerful rainy LFI restoring model based on the adversarial recurrent neural network to progressively eliminate rain streaks and recover the rain-free LFI.

Dongshuo Yin, Leiyi Hu, Bin Li et al.

Pre-training & fine-tuning can enhance the transferring efficiency and performance in visual tasks. Recent delta-tuning methods provide more options for visual classification tasks. Despite their success, existing visual delta-tuning art fails to exceed the upper limit of full fine-tuning on challenging tasks like instance segmentation and semantic segmentation. To find a competitive alternative to full fine-tuning, we propose the Multi-cognitive Visual Adapter (Mona) tuning, a novel adapter-based tuning method. First, we introduce multiple vision-friendly filters into the adapter to enhance its ability to process visual signals, while previous methods mainly rely on language-friendly linear filters. Second, we add the scaled normalization layer in the adapter to regulate the distribution of input features for visual filters. To fully demonstrate the practicality and generality of Mona, we conduct experiments on multiple representative visual tasks, including instance segmentation on COCO, semantic segmentation on ADE20K, object detection on Pascal VOC, and image classification on several common datasets. Exciting results illustrate that Mona surpasses full fine-tuning on all these tasks and is the only delta-tuning method outperforming full fine-tuning on instance segmentation and semantic segmentation tasks. For example, Mona achieves a 1% performance gain on the COCO dataset compared to full fine-tuning. Comprehensive results suggest that Mona-tuning is more suitable for retaining and utilizing the capabilities of pre-trained models than full fine-tuning. The code will be released at https://github.com/Leiyi-Hu/mona.

Yinxuan Huang, Tonglin Chen, Zhimeng Shen et al.

Humans possess the cognitive ability to comprehend scenes in a compositional manner. To empower AI systems with similar capabilities, object-centric learning aims to acquire representations of individual objects from visual scenes without any supervision. Although recent advances in object-centric learning have made remarkable progress on complex synthesis datasets, there is a huge challenge for application to complex real-world scenes. One of the essential reasons is the scarcity of real-world datasets specifically tailored to object-centric learning. To address this problem, we propose a versatile real-world dataset of tabletop scenes for object-centric learning called OCTScenes, which is meticulously designed to serve as a benchmark for comparing, evaluating, and analyzing object-centric learning methods. OCTScenes contains 5000 tabletop scenes with a total of 15 objects. Each scene is captured in 60 frames covering a 360-degree perspective. Consequently, OCTScenes is a versatile benchmark dataset that can simultaneously satisfy the evaluation of object-centric learning methods based on single-image, video, and multi-view. Extensive experiments of representative object-centric learning methods are conducted on OCTScenes. The results demonstrate the shortcomings of state-of-the-art methods for learning meaningful representations from real-world data, despite their impressive performance on complex synthesis datasets. Furthermore, OCTScenes can serve as a catalyst for the advancement of existing methods, inspiring them to adapt to real-world scenes. Dataset and code are available at https://huggingface.co/datasets/Yinxuan/OCTScenes.

Shangchao Su, Bin Li, Chengzhi Zhang et al.

Detection models trained by one party (including server) may face severe performance degradation when distributed to other users (clients). Federated learning can enable multi-party collaborative learning without leaking client data. In this paper, we focus on a special cross-domain scenario in which the server has large-scale labeled data and multiple clients only have a small amount of labeled data; meanwhile, there exist differences in data distributions among the clients. In this case, traditional federated learning methods can't help a client learn both the global knowledge of all participants and its own unique knowledge. To make up for this limitation, we propose a cross-domain federated object detection framework, named FedOD. The proposed framework first performs the federated training to obtain a public global aggregated model through multi-teacher distillation, and sends the aggregated model back to each client for fine-tuning its personalized local model. After a few rounds of communication, on each client we can perform weighted ensemble inference on the public global model and the personalized local model. We establish a federated object detection dataset which has significant background differences and instance differences based on multiple public autonomous driving datasets, and then conduct extensive experiments on the dataset. The experimental results validate the effectiveness of the proposed method.

Jinze Chen, Wei Zhai, Han Han et al.

Event-based vision encodes dynamic scenes as asynchronous spatio-temporal spikes called events. To leverage conventional image processing pipelines, events are typically binned into frames. However, binning functions are discontinuous, which truncates gradients at the frame level and forces most event-based algorithms to rely solely on frame-based features. Attempts to directly learn from raw events avoid this restriction but instead suffer from biased gradient estimation due to the discontinuities of the binning operation, ultimately limiting their learning efficiency. To address this challenge, we propose a novel framework for unbiased gradient estimation of arbitrary binning functions by synthesizing weak derivatives during backpropagation while keeping the forward output unchanged. The key idea is to exploit integration by parts: lifting the target functions to functionals yields an integral form of the derivative of the binning function during backpropagation, where the cotangent function naturally arises. By reconstructing this cotangent function from the sampled cotangent vector, we compute weak derivatives that provably match long-range finite differences of both smooth and non-smooth targets. Experimentally, our method improves simple optimization-based egomotion estimation with 3.2\% lower RMS error and 1.57$\times$ faster convergence. On complex downstream tasks, we achieve 9.4\% lower EPE in self-supervised optical flow, and 5.1\% lower RMS error in SLAM, demonstrating broad benefits for event-based visual perception. Source code can be found at https://github.com/chjz1024/EventFBP.

Shunquan Tan, Qiushi Li, Laiyuan Li et al.

Recent studies shows that the majority of existing deep steganalysis models have a large amount of redundancy, which leads to a huge waste of storage and computing resources. The existing model compression method cannot flexibly compress the convolutional layer in residual shortcut block so that a satisfactory shrinking rate cannot be obtained. In this paper, we propose STD-NET, an unsupervised deep-learning architecture search approach via hierarchical tensor decomposition for image steganalysis. Our proposed strategy will not be restricted by various residual connections, since this strategy does not change the number of input and output channels of the convolution block. We propose a normalized distortion threshold to evaluate the sensitivity of each involved convolutional layer of the base model to guide STD-NET to compress target network in an efficient and unsupervised approach, and obtain two network structures of different shapes with low computation cost and similar performance compared with the original one. Extensive experiments have confirmed that, on one hand, our model can achieve comparable or even better detection performance in various steganalytic scenarios due to the great adaptivity of the obtained network architecture. On the other hand, the experimental results also demonstrate that our proposed strategy is more efficient and can remove more redundancy compared with previous steganalytic network compression methods.

Cong Huang, Jiahao Li, Bin Li et al.

How to efficiently utilize the temporal features is crucial, yet challenging, for video restoration. The temporal features usually contain various noisy and uncorrelated information, and they may interfere with the restoration of the current frame. This paper proposes learning noise-robust feature representations to help video restoration. We are inspired by that the neural codec is a natural denoiser. In neural codec, the noisy and uncorrelated contents which are hard to predict but cost lots of bits are more inclined to be discarded for bitrate saving. Therefore, we design a neural compression module to filter the noise and keep the most useful information in features for video restoration. To achieve robustness to noise, our compression module adopts a spatial channel-wise quantization mechanism to adaptively determine the quantization step size for each position in the latent. Experiments show that our method can significantly boost the performance on video denoising, where we obtain 0.13 dB improvement over BasicVSR++ with only 0.23x FLOPs. Meanwhile, our method also obtains SOTA results on video deraining and dehazing.

Qizao Wang, Xuelin Qian, Bin Li et al.

Person Re-IDentification (Re-ID) as a retrieval task, has achieved tremendous development over the past decade. Existing state-of-the-art methods follow an analogous framework to first extract features from the input images and then categorize them with a classifier. However, since there is no identity overlap between training and testing sets, the classifier is often discarded during inference. Only the extracted features are used for person retrieval via distance metrics. In this paper, we rethink the role of the classifier in person Re-ID, and advocate a new perspective to conceive the classifier as a projection from image features to class prototypes. These prototypes are exactly the learned parameters of the classifier. In this light, we describe the identity of input images as similarities to all prototypes, which are then utilized as more discriminative features to perform person Re-ID. We thereby propose a new baseline ProNet, which innovatively reserves the function of the classifier at the inference stage. To facilitate the learning of class prototypes, both triplet loss and identity classification loss are applied to features that undergo the projection by the classifier. An improved version of ProNet++ is presented by further incorporating multi-granularity designs. Experiments on four benchmarks demonstrate that our proposed ProNet is simple yet effective, and significantly beats previous baselines. ProNet++ also achieves competitive or even better results than transformer-based competitors.

Dong Wei, Huaijiang Sun, Bin Li et al.

Stochastic human motion prediction aims to forecast multiple plausible future motions given a single pose sequence from the past. Most previous works focus on designing elaborate losses to improve the accuracy, while the diversity is typically characterized by randomly sampling a set of latent variables from the latent prior, which is then decoded into possible motions. This joint training of sampling and decoding, however, suffers from posterior collapse as the learned latent variables tend to be ignored by a strong decoder, leading to limited diversity. Alternatively, inspired by the diffusion process in nonequilibrium thermodynamics, we propose MotionDiff, a diffusion probabilistic model to treat the kinematics of human joints as heated particles, which will diffuse from original states to a noise distribution. This process offers a natural way to obtain the "whitened" latents without any trainable parameters, and human motion prediction can be regarded as the reverse diffusion process that converts the noise distribution into realistic future motions conditioned on the observed sequence. Specifically, MotionDiff consists of two parts: a spatial-temporal transformer-based diffusion network to generate diverse yet plausible motions, and a graph convolutional network to further refine the outputs. Experimental results on two datasets demonstrate that our model yields the competitive performance in terms of both accuracy and diversity.

Yixuan Weng, Minjun Zhu, Fei Xia et al.

Language models' (LMs) proficiency in handling deterministic symbolic reasoning and rule-based tasks remains limited due to their dependency implicit learning on textual data. To endow LMs with genuine rule comprehension abilities, we propose "Neural Comprehension" - a framework that synergistically integrates compiled neural networks (CoNNs) into the standard transformer architecture. CoNNs are neural modules designed to explicitly encode rules through artificially generated attention weights. By incorporating CoNN modules, the Neural Comprehension framework enables LMs to accurately and robustly execute rule-intensive symbolic tasks. Extensive experiments demonstrate the superiority of our approach over existing techniques in terms of length generalization, efficiency, and interpretability for symbolic operations. Furthermore, it can be applied to LMs across different model scales, outperforming tool-calling methods in arithmetic reasoning tasks while maintaining superior inference efficiency. Our work highlights the potential of seamlessly unifying explicit rule learning via CoNNs and implicit pattern learning in LMs, paving the way for true symbolic comprehension capabilities.

Jie Wang, Rui Yang, Zijie Geng et al.

Generalization in partially observed markov decision processes (POMDPs) is critical for successful applications of visual reinforcement learning (VRL) in real scenarios. A widely used idea is to learn task-relevant representations that encode task-relevant information of common features in POMDPs, i.e., rewards and transition dynamics. As transition dynamics in the latent state space -- which are task-relevant and invariant to visual distractions -- are unknown to the agents, existing methods alternatively use transition dynamics in the observation space to extract task-relevant information in transition dynamics. However, such transition dynamics in the observation space involve task-irrelevant visual distractions, degrading the generalization performance of VRL methods. To tackle this problem, we propose the reward sequence distribution conditioned on the starting observation and the predefined subsequent action sequence (RSD-OA). The appealing features of RSD-OA include that: (1) RSD-OA is invariant to visual distractions, as it is conditioned on the predefined subsequent action sequence without task-irrelevant information from transition dynamics, and (2) the reward sequence captures long-term task-relevant information in both rewards and transition dynamics. Experiments demonstrate that our representation learning approach based on RSD-OA significantly improves the generalization performance on unseen environments, outperforming several state-of-the-arts on DeepMind Control tasks with visual distractions.

Dongbo Wang, Chang Liu, Zhixiao Zhao et al.

In the context of the rapid development of large language models, we have meticulously trained and introduced the GujiBERT and GujiGPT language models, which are foundational models specifically designed for intelligent information processing of ancient texts. These models have been trained on an extensive dataset that encompasses both simplified and traditional Chinese characters, allowing them to effectively handle various natural language processing tasks related to ancient books, including but not limited to automatic sentence segmentation, punctuation, word segmentation, part-of-speech tagging, entity recognition, and automatic translation. Notably, these models have exhibited exceptional performance across a range of validation tasks using publicly available datasets. Our research findings highlight the efficacy of employing self-supervised methods to further train the models using classical text corpora, thus enhancing their capability to tackle downstream tasks. Moreover, it is worth emphasizing that the choice of font, the scale of the corpus, and the initial model selection all exert significant influence over the ultimate experimental outcomes. To cater to the diverse text processing preferences of researchers in digital humanities and linguistics, we have developed three distinct categories comprising a total of nine model variations. We believe that by sharing these foundational language models specialized in the domain of ancient texts, we can facilitate the intelligent processing and scholarly exploration of ancient literary works and, consequently, contribute to the global dissemination of China's rich and esteemed traditional culture in this new era.

Zhaoyang Jia, Naifu Xue, Zihan Zheng et al.

Recent advanced diffusion methods typically derive strong generative priors by scaling diffusion transformers. However, scaling fails to generalize when adapted for real-time compression scenarios that demand lightweight models. In this paper, we explore the design of real-time and lightweight diffusion codecs by addressing two pivotal questions. First, does diffusion pre-training benefit lightweight diffusion codecs? Through systematic analysis, we find that generation-oriented pre-training is less effective at small model scales whereas compression-oriented pre-training yields consistently better performance. Second, are transformers essential? We find that while global attention is crucial for standard generation, lightweight convolutions suffice for compression-oriented diffusion when paired with distillation. Guided by these findings, we establish a one-step lightweight convolution diffusion codec that achieves real-time $60$~FPS encoding and $42$~FPS decoding at 1080p. Further enhanced by distillation and adversarial learning, the proposed codec reduces bitrate by 85\% at a comparable FID to MS-ILLM, bridging the gap between generative compression and practical real-time deployment. Codes are released at https://github.com/microsoft/GenCodec/tree/main/CoD_Lite

Bin Li, Yixuan Weng, Bin Sun et al.

The temporal answering grounding in the video (TAGV) is a new task naturally derived from temporal sentence grounding in the video (TSGV). Given an untrimmed video and a text question, this task aims at locating the matching span from the video that can semantically answer the question. Existing methods tend to formulate the TAGV task with a visual span-based question answering (QA) approach by matching the visual frame span queried by the text question. However, due to the weak correlations and huge gaps of the semantic features between the textual question and visual answer, existing methods adopting visual span predictor perform poorly in the TAGV task. To bridge these gaps, we propose a visual-prompt text span localizing (VPTSL) method, which introduces the timestamped subtitles as a passage to perform the text span localization for the input text question, and prompts the visual highlight features into the pre-trained language model (PLM) for enhancing the joint semantic representations. Specifically, the context query attention is utilized to perform cross-modal interaction between the extracted textual and visual features. Then, the highlight features are obtained through the video-text highlighting for the visual prompt. To alleviate semantic differences between textual and visual features, we design the text span predictor by encoding the question, the subtitles, and the prompted visual highlight features with the PLM. As a result, the TAGV task is formulated to predict the span of subtitles matching the visual answer. Extensive experiments on the medical instructional dataset, namely MedVidQA, show that the proposed VPTSL outperforms the state-of-the-art (SOTA) method by 28.36% in terms of mIOU with a large margin, which demonstrates the effectiveness of the proposed visual prompt and the text span predictor.

Qingyue Yang, Jie Wang, Xing Li et al.

Attention patterns play a crucial role in both training and inference of large language models (LLMs). Prior works have identified individual patterns such as retrieval heads, sink heads, and diagonal traces, yet these observations remain fragmented and lack a unifying explanation. To bridge this gap, we introduce \textbf{Temporal Attention Pattern Predictability Analysis (TAPPA), a unifying framework that explains diverse attention patterns by analyzing their underlying mathematical formulations} from a temporally continuous perspective. TAPPA both deepens the understanding of attention behavior and guides inference acceleration approaches. Specifically, TAPPA characterizes attention patterns as predictable patterns with clear regularities and unpredictable patterns that appear effectively random. Our analysis further reveals that this distinction can be explained by the degree of query self-similarity along the temporal dimension. Focusing on the predictable patterns, we further provide a detailed mathematical analysis of three representative cases through the joint effect of queries, keys, and Rotary Positional Embeddings (RoPE). We validate TAPPA by applying its insights to KV cache compression and LLM pruning tasks. Across these tasks, a simple metric motivated by TAPPA consistently improves performance over baseline methods. The code is available at https://github.com/MIRALab-USTC/LLM-TAPPA.

Ziqiang Li, Hong Sun, Pengfei Xia et al.

Poisoning efficiency is crucial in poisoning-based backdoor attacks, as attackers aim to minimize the number of poisoning samples while maximizing attack efficacy. Recent studies have sought to enhance poisoning efficiency by selecting effective samples. However, these studies typically rely on a proxy backdoor injection task to identify an efficient set of poisoning samples. This proxy attack-based approach can lead to performance degradation if the proxy attack settings differ from those of the actual victims, due to the shortcut nature of backdoor learning. Furthermore, proxy attack-based methods are extremely time-consuming, as they require numerous complete backdoor injection processes for sample selection. To address these concerns, we present a Proxy attack-Free Strategy (PFS) designed to identify efficient poisoning samples based on the similarity between clean samples and their corresponding poisoning samples, as well as the diversity of the poisoning set. The proposed PFS is motivated by the observation that selecting samples with high similarity between clean and corresponding poisoning samples results in significantly higher attack success rates compared to using samples with low similarity. Additionally, we provide theoretical foundations to explain the proposed PFS. We comprehensively evaluate the proposed strategy across various datasets, triggers, poisoning rates, architectures, and training hyperparameters. Our experimental results demonstrate that PFS enhances backdoor attack efficiency while also offering a remarkable speed advantage over previous proxy attack-based selection methodologies.

Zhiyu Jin, Xuli Shen, Bin Li et al.

Diffusion models (DMs) have recently gained attention with state-of-the-art performance in text-to-image synthesis. Abiding by the tradition in deep learning, DMs are trained and evaluated on the images with fixed sizes. However, users are demanding for various images with specific sizes and various aspect ratio. This paper focuses on adapting text-to-image diffusion models to handle such variety while maintaining visual fidelity. First we observe that, during the synthesis, lower resolution images suffer from incomplete object portrayal, while higher resolution images exhibit repetitively disordered presentation. Next, we establish a statistical relationship indicating that attention entropy changes with token quantity, suggesting that models aggregate spatial information in proportion to image resolution. The subsequent interpretation on our observations is that objects are incompletely depicted due to limited spatial information for low resolutions, while repetitively disorganized presentation arises from redundant spatial information for high resolutions. From this perspective, we propose a scaling factor to alleviate the change of attention entropy and mitigate the defective pattern observed. Extensive experimental results validate the efficacy of the proposed scaling factor, enabling models to achieve better visual effects, image quality, and text alignment. Notably, these improvements are achieved without additional training or fine-tuning techniques.

Kaiqing Lin, Yuzhen Lin, Weixiang Li et al.

The proliferation of deepfake faces poses huge potential negative impacts on our daily lives. Despite substantial advancements in deepfake detection over these years, the generalizability of existing methods against forgeries from unseen datasets or created by emerging generative models remains constrained. In this paper, inspired by the zero-shot advantages of Vision-Language Models (VLMs), we propose a novel approach that repurposes a well-trained VLM for general deepfake detection. Motivated by the model reprogramming paradigm that manipulates the model prediction via input perturbations, our method can reprogram a pre-trained VLM model (e.g., CLIP) solely based on manipulating its input without tuning the inner parameters. First, learnable visual perturbations are used to refine feature extraction for deepfake detection. Then, we exploit information of face embedding to create sample-level adaptative text prompts, improving the performance. Extensive experiments on several popular benchmark datasets demonstrate that (1) the cross-dataset and cross-manipulation performances of deepfake detection can be significantly and consistently improved (e.g., over 88\% AUC in cross-dataset setting from FF++ to WildDeepfake); (2) the superior performances are achieved with fewer trainable parameters, making it a promising approach for real-world applications.

Zhe Chen, Jiannan Wu, Wenhai Wang et al.

The exponential growth of large language models (LLMs) has opened up numerous possibilities for multimodal AGI systems. However, the progress in vision and vision-language foundation models, which are also critical elements of multi-modal AGI, has not kept pace with LLMs. In this work, we design a large-scale vision-language foundation model (InternVL), which scales up the vision foundation model to 6 billion parameters and progressively aligns it with the LLM, using web-scale image-text data from various sources. This model can be broadly applied to and achieve state-of-the-art performance on 32 generic visual-linguistic benchmarks including visual perception tasks such as image-level or pixel-level recognition, vision-language tasks such as zero-shot image/video classification, zero-shot image/video-text retrieval, and link with LLMs to create multi-modal dialogue systems. It has powerful visual capabilities and can be a good alternative to the ViT-22B. We hope that our research could contribute to the development of multi-modal large models. Code and models are available at https://github.com/OpenGVLab/InternVL.

Jiucui Lu, Jiaran Zhou, Junyu Dong et al.

The prominent progress in generative models has significantly improved the reality of generated faces, bringing serious concerns to society. Since recent GAN-generated faces are in high realism, the forgery traces have become more imperceptible, increasing the forensics challenge. To combat GAN-generated faces, many countermeasures based on Convolutional Neural Networks (CNNs) have been spawned due to their strong learning ability. In this paper, we rethink this problem and explore a new approach based on forest models instead of CNNs. Specifically, we describe a simple and effective forest-based method set called {\em ForensicsForest Family} to detect GAN-generate faces. The proposed ForensicsForest family is composed of three variants, which are {\em ForensicsForest}, {\em Hybrid ForensicsForest} and {\em Divide-and-Conquer ForensicsForest} respectively. ForenscisForest is a newly proposed Multi-scale Hierarchical Cascade Forest, which takes semantic, frequency and biology features as input, hierarchically cascades different levels of features for authenticity prediction, and then employs a multi-scale ensemble scheme that can comprehensively consider different levels of information to improve the performance further. Based on ForensicsForest, we develop Hybrid ForensicsForest, an extended version that integrates the CNN layers into models, to further refine the effectiveness of augmented features. Moreover, to reduce the memory cost in training, we propose Divide-and-Conquer ForensicsForest, which can construct a forest model using only a portion of training samplings. In the training stage, we train several candidate forest models using the subsets of training samples. Then a ForensicsForest is assembled by picking the suitable components from these candidate forest models...

Tongzhi Niu, Bin Li, Kai Li et al.

In the surface defect detection, there are some suspicious regions that cannot be uniquely classified as abnormal or normal. The annotating of suspicious regions is easily affected by factors such as workers' emotional fluctuations and judgment standard, resulting in noisy labels, which in turn leads to missing and false detections, and ultimately leads to inconsistent judgments of product quality. Unlike the usual noisy labels, the ones used for surface defect detection appear to be inconsistent rather than mislabeled. The noise occurs in almost every label and is difficult to correct or evaluate. In this paper, we proposed a framework that learns trustworthy models from noisy labels for surface defect defection. At first, to avoid the negative impact of noisy labels on the model, we represent the suspicious regions with consistent and precise elements at the pixel-level and redesign the loss function. Secondly, without changing network structure and adding any extra labels, pluggable spatially correlated Bayesian module is proposed. Finally, the defect discrimination confidence is proposed to measure the uncertainty, with which anomalies can be identified as defects. Our results indicate not only the effectiveness of the proposed method in learning from noisy labels, but also robustness and real-time performance.

Miao Wang, Yuling Shi, Yijiang Li et al.

Text-based role-playing models can imitate character styles, yet they often fail to reflect a scene's atmosphere and evolving tension, both essential for immersive applications such as Virtual Reality (VR) games and interactive narratives. We study video-grounded role-playing dialogue and introduce EBM-RL (Eye-Brain-Mouth Reinforcement Learning), a decoupled GRPO-based framework that explicitly separates observation ([perception]), reasoning ([think]), and utterance ([answer]). This structure promotes human-like sensory grounding by compelling the model to first attend to visual cues, then form internal interpretations, and finally generate context-appropriate dialogue. EBM-RL integrates four complementary rewards: (i) CLIP-based scene-text alignment to improve ambiance and emotion; (ii) a Perceptual-Cognitive reward that encourages [perception] and [think] processes that increase the likelihood of the reference response; (iii) answer accuracy to ensure faithfulness; and (iv) a dense format reward to enforce the desired structured output. Extensive experiments demonstrate that EBM-RL substantially outperforms text-only role-playing baselines and larger-scale vision-language models on our immersive role-playing benchmark, delivering simultaneous gains in visual-atmosphere consistency and character authenticity. Beyond the role-playing domain, EBM-RL also exhibits strong zero-shot generalization: without any additional fine-tuning, it consistently improves performance on out-of-domain VideoQA benchmarks. We additionally release an open-source dataset for video-grounded role-playing dialogue.