Zhen Zhu

Zhen Zhu, Yiming Gong, Derek Hoiem

We propose an approach for anytime continual learning (AnytimeCL) for open vocabulary image classification. The AnytimeCL problem aims to break away from batch training and rigid models by requiring that a system can predict any set of labels at any time and efficiently update and improve when receiving one or more training samples at any time. Despite the challenging goal, we achieve substantial improvements over recent methods. We propose a dynamic weighting between predictions of a partially fine-tuned model and a fixed open vocabulary model that enables continual improvement when training samples are available for a subset of a task's labels. We also propose an attention-weighted PCA compression of training features that reduces storage and computation with little impact to model accuracy. Our methods are validated with experiments that test flexibility of learning and inference. Code is available at https://github.com/jessemelpolio/AnytimeCL.

Zhen Zhu, Weijie Lyu, Yao Xiao et al.

We introduce a method for flexible and efficient continual learning in open-vocabulary image classification, drawing inspiration from the complementary learning systems observed in human cognition. Specifically, we propose to combine predictions from a CLIP zero-shot model and the exemplar-based model, using the zero-shot estimated probability that a sample's class is within the exemplar classes. We also propose a "tree probe" method, an adaption of lazy learning principles, which enables fast learning from new examples with competitive accuracy to batch-trained linear models. We test in data incremental, class incremental, and task incremental settings, as well as ability to perform flexible inference on varying subsets of zero-shot and learned categories. Our proposed method achieves a good balance of learning speed, target task effectiveness, and zero-shot effectiveness. Code will be available at https://github.com/jessemelpolio/TreeProbe.

Kimi Team, Tongtong Bai, Yifan Bai et al.

We introduce Kimi K2.5, an open-source multimodal agentic model designed to advance general agentic intelligence. K2.5 emphasizes the joint optimization of text and vision so that two modalities enhance each other. This includes a series of techniques such as joint text-vision pre-training, zero-vision SFT, and joint text-vision reinforcement learning. Building on this multimodal foundation, K2.5 introduces Agent Swarm, a self-directed parallel agent orchestration framework that dynamically decomposes complex tasks into heterogeneous sub-problems and executes them concurrently. Extensive evaluations show that Kimi K2.5 achieves state-of-the-art results across various domains including coding, vision, reasoning, and agentic tasks. Agent Swarm also reduces latency by up to $4.5\times$ over single-agent baselines. We release the post-trained Kimi K2.5 model checkpoint to facilitate future research and real-world applications of agentic intelligence.

Zhen Zhu, Yijun Li, Weijie Lyu et al.

We investigate how to generate multimodal image outputs, such as RGB, depth, and surface normals, with a single generative model. The challenge is to produce outputs that are realistic, and also consistent with each other. Our solution builds on the StyleGAN3 architecture, with a shared backbone and modality-specific branches in the last layers of the synthesis network, and we propose per-modality fidelity discriminators and a cross-modality consistency discriminator. In experiments on the Stanford2D3D dataset, we demonstrate realistic and consistent generation of RGB, depth, and normal images. We also show a training recipe to easily extend our pretrained model on a new domain, even with a few pairwise data. We further evaluate the use of synthetically generated RGB and depth pairs for training or fine-tuning depth estimators. Code will be available at https://github.com/jessemelpolio/MultimodalGAN.

Zijie Wu, Zhen Zhu, Junping Du et al.

In this paper, we aim to devise a universally versatile style transfer method capable of performing artistic, photo-realistic, and video style transfer jointly, without seeing videos during training. Previous single-frame methods assume a strong constraint on the whole image to maintain temporal consistency, which could be violated in many cases. Instead, we make a mild and reasonable assumption that global inconsistency is dominated by local inconsistencies and devise a generic Contrastive Coherence Preserving Loss (CCPL) applied to local patches. CCPL can preserve the coherence of the content source during style transfer without degrading stylization. Moreover, it owns a neighbor-regulating mechanism, resulting in a vast reduction of local distortions and considerable visual quality improvement. Aside from its superior performance on versatile style transfer, it can be easily extended to other tasks, such as image-to-image translation. Besides, to better fuse content and style features, we propose Simple Covariance Transformation (SCT) to effectively align second-order statistics of the content feature with the style feature. Experiments demonstrate the effectiveness of the resulting model for versatile style transfer, when armed with CCPL.

Zijie Wu, Chaohui Yu, Zhen Zhu et al.

Recent progress in text-to-image (T2I) models enables high-quality image generation with flexible textual control. To utilize the abundant visual priors in the off-the-shelf T2I models, a series of methods try to invert an image to proper embedding that aligns with the semantic space of the T2I model. However, these image-to-text (I2T) inversion methods typically need multiple source images containing the same concept or struggle with the imbalance between editing flexibility and visual fidelity. In this work, we point out that the critical problem lies in the foreground-background entanglement when learning an intended concept, and propose a simple and effective baseline for single-image I2T inversion, named SingleInsert. SingleInsert adopts a two-stage scheme. In the first stage, we regulate the learned embedding to concentrate on the foreground area without being associated with the irrelevant background. In the second stage, we finetune the T2I model for better visual resemblance and devise a semantic loss to prevent the language drift problem. With the proposed techniques, SingleInsert excels in single concept generation with high visual fidelity while allowing flexible editing. Additionally, SingleInsert can perform single-image novel view synthesis and multiple concepts composition without requiring joint training. To facilitate evaluation, we design an editing prompt list and introduce a metric named Editing Success Rate (ESR) for quantitative assessment of editing flexibility. Our project page is: https://jarrentwu1031.github.io/SingleInsert-web/

Guy Kaplan, Zorik Gekhman, Zhen Zhu et al.

Large language models are prone to hallucinating factually incorrect statements. A key source of these errors is exposure to new factual information through supervised fine-tuning (SFT), which can increase hallucinations w.r.t. knowledge acquired during pre-training. In this work, we explore whether SFT-induced hallucinations can be mitigated using established tools from the continual learning literature, since they arise as a by-product of knowledge degradation during training. We propose a self-distillation-based SFT method that facilitates effective factual learning while minimizing hallucinations w.r.t. pre-existing knowledge by regularizing output-distribution drift. We also show that, in settings where new knowledge acquisition is unnecessary, suppressing factual plasticity by freezing parameter groups, can preserve task performance while reducing hallucinations. Lastly, we investigate the mechanism behind SFT-induced hallucinations through three hypotheses: capacity limitations, behavior cloning, and localized interference. Our experiments show that a main driver is interference among overlapping semantic representations, and that self-distillation succeeds by mitigating this interference.

Kimi Team, Yifan Bai, Yiping Bao et al. · tsinghua

We introduce Kimi K2, a Mixture-of-Experts (MoE) large language model with 32 billion activated parameters and 1 trillion total parameters. We propose the MuonClip optimizer, which improves upon Muon with a novel QK-clip technique to address training instability while enjoying the advanced token efficiency of Muon. Based on MuonClip, K2 was pre-trained on 15.5 trillion tokens with zero loss spike. During post-training, K2 undergoes a multi-stage post-training process, highlighted by a large-scale agentic data synthesis pipeline and a joint reinforcement learning (RL) stage, where the model improves its capabilities through interactions with real and synthetic environments. Kimi K2 achieves state-of-the-art performance among open-source non-thinking models, with strengths in agentic capabilities. Notably, K2 obtains 66.1 on Tau2-Bench, 76.5 on ACEBench (En), 65.8 on SWE-Bench Verified, and 47.3 on SWE-Bench Multilingual -- surpassing most open and closed-sourced baselines in non-thinking settings. It also exhibits strong capabilities in coding, mathematics, and reasoning tasks, with a score of 53.7 on LiveCodeBench v6, 49.5 on AIME 2025, 75.1 on GPQA-Diamond, and 27.1 on OJBench, all without extended thinking. These results position Kimi K2 as one of the most capable open-source large language models to date, particularly in software engineering and agentic tasks. We release our base and post-trained model checkpoints to facilitate future research and applications of agentic intelligence.

Kimi Team, Angang Du, Bofei Gao et al. · pku, tsinghua

Language model pretraining with next token prediction has proved effective for scaling compute but is limited to the amount of available training data. Scaling reinforcement learning (RL) unlocks a new axis for the continued improvement of artificial intelligence, with the promise that large language models (LLMs) can scale their training data by learning to explore with rewards. However, prior published work has not produced competitive results. In light of this, we report on the training practice of Kimi k1.5, our latest multi-modal LLM trained with RL, including its RL training techniques, multi-modal data recipes, and infrastructure optimization. Long context scaling and improved policy optimization methods are key ingredients of our approach, which establishes a simplistic, effective RL framework without relying on more complex techniques such as Monte Carlo tree search, value functions, and process reward models. Notably, our system achieves state-of-the-art reasoning performance across multiple benchmarks and modalities -- e.g., 77.5 on AIME, 96.2 on MATH 500, 94-th percentile on Codeforces, 74.9 on MathVista -- matching OpenAI's o1. Moreover, we present effective long2short methods that use long-CoT techniques to improve short-CoT models, yielding state-of-the-art short-CoT reasoning results -- e.g., 60.8 on AIME, 94.6 on MATH500, 47.3 on LiveCodeBench -- outperforming existing short-CoT models such as GPT-4o and Claude Sonnet 3.5 by a large margin (up to +550%).

Zhen Zhu, Yiming Gong, Yao Xiao et al.

How can we teach large multimodal models (LMMs) new skills without erasing prior abilities? We study sequential fine-tuning on five target skills while monitoring general ability on eight held-out benchmarks across three model families. We observe that apparent "forgetting" on held-out tasks after narrow fine-tuning can partly recover at later stages. We trace this behavior to a measurable shift in the output token distribution, manifested through a simple counting-bias probe that co-varies with forgetting. Guided by this picture, we identify two simple, robust tuning recipes that learn strongly while limiting drift: (i) updating only the self-attention projection layers, and (ii) updating only the MLP Gate&Up while freezing the Down projection. Across models and tasks, these choices deliver strong target gains while largely preserving held-out performance. Code is available at https://github.com/jessemelpolio/LMM_CL

Maosheng Qin, Renyu Zhu, Mingxuan Xia et al.

High-quality annotated data is a cornerstone of modern Natural Language Processing (NLP). While recent methods begin to leverage diverse annotation sources-including Large Language Models (LLMs), Small Language Models (SLMs), and human experts-they often focus narrowly on the labeling step itself. A critical gap remains in the holistic process control required to manage these sources dynamically, addressing complex scheduling and quality-cost trade-offs in a unified manner. Inspired by real-world crowdsourcing companies, we introduce CrowdAgent, a multi-agent system that provides end-to-end process control by integrating task assignment, data annotation, and quality/cost management. It implements a novel methodology that rationally assigns tasks, enabling LLMs, SLMs, and human experts to advance synergistically in a collaborative annotation workflow. We demonstrate the effectiveness of CrowdAgent through extensive experiments on six diverse multimodal classification tasks. The source code and video demo are available at https://github.com/QMMMS/CrowdAgent.

Hanshen Zhu, Zhen Zhu, Kaile Zhang et al.

We tackle the task of geometric image editing, where an object within an image is repositioned, reoriented, or reshaped while preserving overall scene coherence. Previous diffusion-based editing methods often attempt to handle all relevant subtasks in a single step, proving difficult when transformations become large or structurally complex. We address this by proposing a decoupled pipeline that separates object transformation, source region inpainting, and target region refinement. Both inpainting and refinement are implemented using a training-free diffusion approach, FreeFine. In experiments on our new GeoBench benchmark, which contains both 2D and 3D editing scenarios, FreeFine outperforms state-of-the-art alternatives in image fidelity, and edit precision, especially under demanding transformations. Code and benchmark are available at: https://github.com/CIawevy/FreeFine

Yao Xiao, Qiqian Fu, Heyi Tao et al.

Image-text models excel at image-level tasks but struggle with detailed visual understanding. While these models provide strong visual-language alignment, segmentation models like SAM2 offer precise spatial boundaries for objects. To this end, we propose TextRegion, a simple, effective, and training-free framework that combines the strengths of image-text models and SAM2 to generate powerful text-aligned region tokens. These tokens enable detailed visual understanding while preserving open-vocabulary capabilities. They can be directly applied to various downstream tasks, including open-world semantic segmentation, referring expression comprehension, and grounding. We conduct extensive evaluations and consistently achieve superior or competitive performance compared to state-of-the-art training-free methods. Additionally, our framework is compatible with many image-text models, making it highly practical and easily extensible as stronger models emerge. Code is available at: https://github.com/avaxiao/TextRegion.

Zhen Zhu, Tengteng Huang, Mengde Xu et al.

This paper proposes a new generative adversarial network for pose transfer, i.e., transferring the pose of a given person to a target pose. We design a progressive generator which comprises a sequence of transfer blocks. Each block performs an intermediate transfer step by modeling the relationship between the condition and the target poses with attention mechanism. Two types of blocks are introduced, namely Pose-Attentional Transfer Block (PATB) and Aligned Pose-Attentional Transfer Bloc ~(APATB). Compared with previous works, our model generates more photorealistic person images that retain better appearance consistency and shape consistency compared with input images. We verify the efficacy of the model on the Market-1501 and DeepFashion datasets, using quantitative and qualitative measures. Furthermore, we show that our method can be used for data augmentation for the person re-identification task, alleviating the issue of data insufficiency. Code and pretrained models are available at https://github.com/tengteng95/Pose-Transfer.git.

Yang Liu, Zhen Zhu, Xiang Bai

Visible watermarks are widely-used in images to protect copyright ownership. Analyzing watermark removal helps to reinforce the anti-attack techniques in an adversarial way. Current removal methods normally leverage image-to-image translation techniques. Nevertheless, the uncertainty of the size, shape, color and transparency of the watermarks set a huge barrier for these methods. To combat this, we combine traditional watermarked image decomposition into a two-stage generator, called Watermark-Decomposition Network (WDNet), where the first stage predicts a rough decomposition from the whole watermarked image and the second stage specifically centers on the watermarked area to refine the removal results. The decomposition formulation enables WDNet to separate watermarks from the images rather than simply removing them. We further show that these separated watermarks can serve as extra nutrients for building a larger training dataset and further improving removal performance. Besides, we construct a large-scale dataset named CLWD, which mainly contains colored watermarks, to fill the vacuum of colored watermark removal dataset. Extensive experiments on the public gray-scale dataset LVW and CLWD consistently show that the proposed WDNet outperforms the state-of-the-art approaches both in accuracy and efficiency. The code and CLWD dataset are publicly available at https://github.com/MRUIL/WDNet.

Zhen Zhu, Zhiliang Xu, Ansheng You et al.

In this paper, we focus on semantically multi-modal image synthesis (SMIS) task, namely, generating multi-modal images at the semantic level. Previous work seeks to use multiple class-specific generators, constraining its usage in datasets with a small number of classes. We instead propose a novel Group Decreasing Network (GroupDNet) that leverages group convolutions in the generator and progressively decreases the group numbers of the convolutions in the decoder. Consequently, GroupDNet is armed with much more controllability on translating semantic labels to natural images and has plausible high-quality yields for datasets with many classes. Experiments on several challenging datasets demonstrate the superiority of GroupDNet on performing the SMIS task. We also show that GroupDNet is capable of performing a wide range of interesting synthesis applications. Codes and models are available at: https://github.com/Seanseattle/SMIS.

Xiangtai Li, Ansheng You, Zhen Zhu et al.

In this paper, we focus on designing effective method for fast and accurate scene parsing. A common practice to improve the performance is to attain high resolution feature maps with strong semantic representation. Two strategies are widely used -- atrous convolutions and feature pyramid fusion, are either computation intensive or ineffective. Inspired by the Optical Flow for motion alignment between adjacent video frames, we propose a Flow Alignment Module (FAM) to learn Semantic Flow between feature maps of adjacent levels, and broadcast high-level features to high resolution features effectively and efficiently. Furthermore, integrating our module to a common feature pyramid structure exhibits superior performance over other real-time methods even on light-weight backbone networks, such as ResNet-18. Extensive experiments are conducted on several challenging datasets, including Cityscapes, PASCAL Context, ADE20K and CamVid. Especially, our network is the first to achieve 80.4\% mIoU on Cityscapes with a frame rate of 26 FPS. The code is available at \url{https://github.com/lxtGH/SFSegNets}.

Zhen Zhu, Mengde Xu, Song Bai et al.

The non-local module works as a particularly useful technique for semantic segmentation while criticized for its prohibitive computation and GPU memory occupation. In this paper, we present Asymmetric Non-local Neural Network to semantic segmentation, which has two prominent components: Asymmetric Pyramid Non-local Block (APNB) and Asymmetric Fusion Non-local Block (AFNB). APNB leverages a pyramid sampling module into the non-local block to largely reduce the computation and memory consumption without sacrificing the performance. AFNB is adapted from APNB to fuse the features of different levels under a sufficient consideration of long range dependencies and thus considerably improves the performance. Extensive experiments on semantic segmentation benchmarks demonstrate the effectiveness and efficiency of our work. In particular, we report the state-of-the-art performance of 81.3 mIoU on the Cityscapes test set. For a 256x128 input, APNB is around 6 times faster than a non-local block on GPU while 28 times smaller in GPU running memory occupation. Code is available at: https://github.com/MendelXu/ANN.git.

Zhen Zhu, Tengteng Huang, Baoguang Shi et al.

This paper proposes a new generative adversarial network for pose transfer, i.e., transferring the pose of a given person to a target pose. The generator of the network comprises a sequence of Pose-Attentional Transfer Blocks that each transfers certain regions it attends to, generating the person image progressively. Compared with those in previous works, our generated person images possess better appearance consistency and shape consistency with the input images, thus significantly more realistic-looking. The efficacy and efficiency of the proposed network are validated both qualitatively and quantitatively on Market-1501 and DeepFashion. Furthermore, the proposed architecture can generate training images for person re-identification, alleviating data insufficiency. Codes and models are available at: https://github.com/tengteng95/Pose-Transfer.git.

Yifei Yao, Wentao He, Chenyu Gu et al.

Training and deploying reinforcement learning (RL) policies for robots, especially in accomplishing specific tasks, presents substantial challenges. Recent advancements have explored diverse reward function designs, training techniques, simulation-to-reality (sim-to-real) transfers, and performance analysis methodologies, yet these still require significant human intervention. This paper introduces an end-to-end framework for training and deploying RL policies, guided by Large Language Models (LLMs), and evaluates its effectiveness on bipedal robots. The framework consists of three interconnected modules: an LLM-guided reward function design module, an RL training module leveraging prior work, and a sim-to-real homomorphic evaluation module. This design significantly reduces the need for human input by utilizing only essential simulation and deployment platforms, with the option to incorporate human-engineered strategies and historical data. We detail the construction of these modules, their advantages over traditional approaches, and demonstrate the framework's capability to autonomously develop and refine controlling strategies for bipedal robot locomotion, showcasing its potential to operate independently of human intervention.

Savya Khosla, Zhen Zhu, Yifei He

This paper explores Memory-Augmented Neural Networks (MANNs), delving into how they blend human-like memory processes into AI. It covers different memory types, like sensory, short-term, and long-term memory, linking psychological theories with AI applications. The study investigates advanced architectures such as Hopfield Networks, Neural Turing Machines, Correlation Matrix Memories, Memformer, and Neural Attention Memory, explaining how they work and where they excel. It dives into real-world uses of MANNs across Natural Language Processing, Computer Vision, Multimodal Learning, and Retrieval Models, showing how memory boosters enhance accuracy, efficiency, and reliability in AI tasks. Overall, this survey provides a comprehensive view of MANNs, offering insights for future research in memory-based AI systems.

Joshua Cho, Sara Aghajanzadeh, Zhen Zhu et al.

In this paper, we present a simple yet highly effective "free lunch" solution for low-light image enhancement (LLIE), which aims to restore low-light images as if acquired in well-illuminated environments. Our method necessitates no optimization, training, fine-tuning, text conditioning, or hyperparameter adjustments, yet it consistently reconstructs low-light images with superior fidelity. Specifically, we leverage a pre-trained text-to-image diffusion prior, learned from training on a large collection of natural images, and the features present in the model itself to guide the inference, in contrast to existing methods that depend on customized constraints. Comprehensive quantitative evaluations demonstrate that our approach outperforms SOTA methods on established datasets, while qualitative analyses indicate enhanced color accuracy and the rectification of subtle chromatic deviations. Furthermore, additional experiments reveal that our method, without any modifications, achieves SOTA-comparable performance in the auto white balance (AWB) task.

Yang Liu, Xiaomin Yu, Gongyu Zhang et al.

"A data scientist is tasked with developing a low-cost surgical VQA system for a 2-month workshop. Due to data sensitivity, she collects 50 hours of surgical video from a hospital, requiring two months for privacy approvals. Privacy restrictions prevent uploading data to platforms like ChatGPT, so she assembles one annotator and a medical expert to manually create QA pairs. This process takes three weeks and costs over $10,000. The trained model provides accurate responses within the limited data scope but lacks broader generalizability, completing the project in 3 months." To simplify the challenges presented in the scenario above. In this paper, we replace the image input with text for Vision-language training. Inspired by prior noise injection methods to reduce modality gaps, we introduce Adaptive ranged cosine Similarity injected noise (ArcSin). First, we introduce an innovative adaptive noise scale that effectively generates the textual elements with more variability while preserving the original text feature's integrity. Second, a similarity pool strategy is employed, expanding the domain generalization potential by broadening the overall noise scale. This dual strategy effectively broadens the scope of the original domain while safeguarding content integrity. Our empirical results demonstrate that these models closely rival those trained on images in terms of performance. Specifically, our method exhibits substantial improvements over the previous state-of-the-art, achieving gains of 1.9 and 1.1 CIDEr points in S-Cap and M-Cap, respectively. Additionally, we observe increases of 0.5 percentage points (pp), 1.4 pp, and 1.4 pp in accuracy for VQA, VQA-E, and VE, respectively, pushing the boundaries of what is achievable within the constraints of image-trained model benchmarks.

Yiming Gong, Zhen Zhu, Minjia Zhang

We propose a fast text-guided image editing method called InstantEdit based on the RectifiedFlow framework, which is structured as a few-step editing process that preserves critical content while following closely to textual instructions. Our approach leverages the straight sampling trajectories of RectifiedFlow by introducing a specialized inversion strategy called PerRFI. To maintain consistent while editable results for RectifiedFlow model, we further propose a novel regeneration method, Inversion Latent Injection, which effectively reuses latent information obtained during inversion to facilitate more coherent and detailed regeneration. Additionally, we propose a Disentangled Prompt Guidance technique to balance editability with detail preservation, and integrate a Canny-conditioned ControlNet to incorporate structural cues and suppress artifacts. Evaluation on the PIE image editing dataset demonstrates that InstantEdit is not only fast but also achieves better qualitative and quantitative results compared to state-of-the-art few-step editing methods.

Xudong Xie, Zhen Zhu, Zijie Wu et al.

This paper aims for a new generation task: non-stationary multi-texture synthesis, which unifies synthesizing multiple non-stationary textures in a single model. Most non-stationary textures have large scale variance and can hardly be synthesized through one model. To combat this, we propose a multi-scale generator to capture structural patterns of various scales and effectively synthesize textures with a minor cost. However, it is still hard to handle textures of different categories with different texture patterns. Therefore, we present a category-specific training strategy to focus on learning texture pattern of a specific domain. Interestingly, once trained, our model is able to produce multi-pattern generations with dynamic variations without the need to finetune the model for different styles. Moreover, an objective evaluation metric is designed for evaluating the quality of texture expansion and global structure consistency. To our knowledge, ours is the first scheme for this challenging task, including model, training, and evaluation. Experimental results demonstrate the proposed method achieves superior performance and time efficiency. The code will be available after the publication.

Zhiliang Xu, Zhibin Hong, Changxing Ding et al.

Advanced face swapping methods have achieved appealing results. However, most of these methods have many parameters and computations, which makes it challenging to apply them in real-time applications or deploy them on edge devices like mobile phones. In this work, we propose a lightweight Identity-aware Dynamic Network (IDN) for subject-agnostic face swapping by dynamically adjusting the model parameters according to the identity information. In particular, we design an efficient Identity Injection Module (IIM) by introducing two dynamic neural network techniques, including the weights prediction and weights modulation. Once the IDN is updated, it can be applied to swap faces given any target image or video. The presented IDN contains only 0.50M parameters and needs 0.33G FLOPs per frame, making it capable for real-time video face swapping on mobile phones. In addition, we introduce a knowledge distillation-based method for stable training, and a loss reweighting module is employed to obtain better synthesized results. Finally, our method achieves comparable results with the teacher models and other state-of-the-art methods.

Zhiliang Xu, Xiyu Yu, Zhibin Hong et al.

Face attribute editing aims to generate faces with one or multiple desired face attributes manipulated while other details are preserved. Unlike prior works such as GAN inversion, which has an expensive reverse mapping process, we propose a simple feed-forward network to generate high-fidelity manipulated faces. By simply employing some existing and easy-obtainable prior information, our method can control, transfer, and edit diverse attributes of faces in the wild. The proposed method can consequently be applied to various applications such as face swapping, face relighting, and makeup transfer. In our method, we decouple identity, expression, pose, and illumination using 3D priors; separate texture and colors by using region-wise style codes. All the information is embedded into adversarial learning by our identity-style normalization module. Disentanglement losses are proposed to enhance the generator to extract information independently from each attribute. Comprehensive quantitative and qualitative evaluations have been conducted. In a single framework, our method achieves the best or competitive scores on a variety of face applications.

Mucun Tian, Chun Guo, Vito Ostuni et al.

Learning-to-Rank (LTR) models trained from implicit feedback (e.g. clicks) suffer from inherent biases. A well-known one is the position bias -- documents in top positions are more likely to receive clicks due in part to their position advantages. To unbiasedly learn to rank, existing counterfactual frameworks first estimate the propensity (probability) of missing clicks with intervention data from a small portion of search traffic, and then use inverse propensity score (IPS) to debias LTR algorithms on the whole data set. These approaches often assume the propensity only depends on the position of the document, which may cause high estimation variance in applications where the search context (e.g. query, user) varies frequently. While context-dependent propensity models reduce variance, accurate estimations may require randomization or intervention on a large amount of traffic, which may not be realistic in real-world systems, especially for long tail queries. In this work, we employ heterogeneous treatment effect estimation techniques to estimate position bias when intervention click data is limited. We then use such estimations to debias the observed click distribution and re-draw a new de-biased data set, which can be used for any LTR algorithms. We conduct simulations with varying experiment conditions and show the effectiveness of the proposed method in regimes with long tail queries and sparse clicks.

Yang Zhou, Zhen Zhu, Xiang Bai et al.

The real world exhibits an abundance of non-stationary textures. Examples include textures with large-scale structures, as well as spatially variant and inhomogeneous textures. While existing example-based texture synthesis methods can cope well with stationary textures, non-stationary textures still pose a considerable challenge, which remains unresolved. In this paper, we propose a new approach for example-based non-stationary texture synthesis. Our approach uses a generative adversarial network (GAN), trained to double the spatial extent of texture blocks extracted from a specific texture exemplar. Once trained, the fully convolutional generator is able to expand the size of the entire exemplar, as well as of any of its sub-blocks. We demonstrate that this conceptually simple approach is highly effective for capturing large-scale structures, as well as other non-stationary attributes of the input exemplar. As a result, it can cope with challenging textures, which, to our knowledge, no other existing method can handle.

Minghui Liao, Zhen Zhu, Baoguang Shi et al.

Text in natural images is of arbitrary orientations, requiring detection in terms of oriented bounding boxes. Normally, a multi-oriented text detector often involves two key tasks: 1) text presence detection, which is a classification problem disregarding text orientation; 2) oriented bounding box regression, which concerns about text orientation. Previous methods rely on shared features for both tasks, resulting in degraded performance due to the incompatibility of the two tasks. To address this issue, we propose to perform classification and regression on features of different characteristics, extracted by two network branches of different designs. Concretely, the regression branch extracts rotation-sensitive features by actively rotating the convolutional filters, while the classification branch extracts rotation-invariant features by pooling the rotation-sensitive features. The proposed method named Rotation-sensitive Regression Detector (RRD) achieves state-of-the-art performance on three oriented scene text benchmark datasets, including ICDAR 2015, MSRA-TD500, RCTW-17 and COCO-Text. Furthermore, RRD achieves a significant improvement on a ship collection dataset, demonstrating its generality on oriented object detection.

Antoine Dedieu, Rahul Mazumder, Zhen Zhu et al.

An important metric of users' satisfaction and engagement within on-line streaming services is the user session length, i.e. the amount of time they spend on a service continuously without interruption. Being able to predict this value directly benefits the recommendation and ad pacing contexts in music and video streaming services. Recent research has shown that predicting the exact amount of time spent is highly nontrivial due to many external factors for which a user can end a session, and the lack of predictive covariates. Most of the other related literature on duration based user engagement has focused on dwell time for websites, for search and display ads, mainly for post-click satisfaction prediction or ad ranking. In this work we present a novel framework inspired by hierarchical Bayesian modeling to predict, at the moment of login, the amount of time a user will spend in the streaming service. The time spent by a user on a platform depends upon user-specific latent variables which are learned via hierarchical shrinkage. Our framework enjoys theoretical guarantees and naturally incorporates flexible parametric/nonparametric models on the covariates, including models robust to outliers. Our proposal is found to outperform state-of- the-art estimators in terms of efficiency and predictive performance on real world public and private datasets.

Gui-Song Xia, Xiang Bai, Jian Ding et al.

Object detection is an important and challenging problem in computer vision. Although the past decade has witnessed major advances in object detection in natural scenes, such successes have been slow to aerial imagery, not only because of the huge variation in the scale, orientation and shape of the object instances on the earth's surface, but also due to the scarcity of well-annotated datasets of objects in aerial scenes. To advance object detection research in Earth Vision, also known as Earth Observation and Remote Sensing, we introduce a large-scale Dataset for Object deTection in Aerial images (DOTA). To this end, we collect $2806$ aerial images from different sensors and platforms. Each image is of the size about 4000-by-4000 pixels and contains objects exhibiting a wide variety of scales, orientations, and shapes. These DOTA images are then annotated by experts in aerial image interpretation using $15$ common object categories. The fully annotated DOTA images contains $188,282$ instances, each of which is labeled by an arbitrary (8 d.o.f.) quadrilateral To build a baseline for object detection in Earth Vision, we evaluate state-of-the-art object detection algorithms on DOTA. Experiments demonstrate that DOTA well represents real Earth Vision applications and are quite challenging.

Pengyuan Lyu, Xiang Bai, Cong Yao et al.

In this paper, we investigate the Chinese calligraphy synthesis problem: synthesizing Chinese calligraphy images with specified style from standard font(eg. Hei font) images (Fig. 1(a)). Recent works mostly follow the stroke extraction and assemble pipeline which is complex in the process and limited by the effect of stroke extraction. We treat the calligraphy synthesis problem as an image-to-image translation problem and propose a deep neural network based model which can generate calligraphy images from standard font images directly. Besides, we also construct a large scale benchmark that contains various styles for Chinese calligraphy synthesis. We evaluate our method as well as some baseline methods on the proposed dataset, and the experimental results demonstrate the effectiveness of our proposed model.

Yuan Sun, Zhen Zhu

Person knowledge extraction is the foundation of the Tibetan knowledge graph construction, which provides support for Tibetan question answering system, information retrieval, information extraction and other researches, and promotes national unity and social stability. This paper proposes a SVM and template based approach to Tibetan person knowledge extraction. Through constructing the training corpus, we build the templates based the shallow parsing analysis of Tibetan syntactic, semantic features and verbs. Using the training corpus, we design a hierarchical SVM classifier to realize the entity knowledge extraction. Finally, experimental results prove the method has greater improvement in Tibetan person knowledge extraction.