Diffusion models have revolutionized text-to-image (T2I) synthesis, producing high-quality, photorealistic images. However, they still struggle to properly render the spatial relationships described in text prompts. To address the lack of spatial information in T2I generations, existing methods typically use external network conditioning and predefined layouts, resulting in higher computational costs and reduced flexibility. Our approach builds upon a curated dataset of spatially explicit prompts, meticulously extracted and synthesized from LAION-400M to ensure precise alignment between textual descriptions and spatial layouts. Alongside this dataset, we present ESPLoRA, a flexible fine-tuning framework based on Low-Rank Adaptation, specifically designed to enhance spatial consistency in generative models without increasing generation time or compromising the quality of the outputs. In addition to ESPLoRA, we propose refined evaluation metrics grounded in geometric constraints, capturing 3D spatial relations such as "in front of" or "behind". These metrics also expose spatial biases in T2I models which, even when not fully mitigated, can be strategically exploited by our TORE algorithm to further improve the spatial consistency of generated images. Our method outperforms CoMPaSS, the current baseline framework, on spatial consistency benchmarks.

Visual intelligence seeks to perceive, interpret, and synthesize the visual world and is central to modern computer vision. Human-centered visual intelligence is especially demanding because it studies people as expressive, socially situated subjects whose meaning is rarely conveyed by appearance alone. It couples vision with audio and language across four representative tasks: human emotion recognition, human video generation, human voice cloning, and human video matting. Yet existing resources remain task-specific, providing modalities and annotations for individual problems rather than a shared foundation coordinating understanding and generation. This limits multimodal signal use and broader research. We address this gap with HUG-VIS, a unified benchmark for Human-centered Understanding and Generation in Visual Intelligence. It contains 8,400 seated half-body videos of 30 professional actors, each performing the same 280 emotion-action-prompt assignments under a controlled Mandarin studio protocol, with synchronized video, audio, text, and alpha mattes. We evaluate diverse open- and closed-source models across the four tasks under a unified zero-shot protocol using automatic metrics, criterion-specific mean opinion scores, and multiple cross-task analyses. Results show that (i) linguistic content dominates current emotion recognition, while purely visual affect recognition is weakest; (ii) in video generation and voice cloning, automatic metrics and human judgment agree overall but differ in their top rankings, requiring joint reporting; (iii) boundary fidelity under motion is the main remaining obstacle for human matting; and (iv) task difficulty varies across emotions, models, and metrics, with notable cross-task correlations. The dataset and results are available at https://github.com/GML-MMGroup/HUG-VIS.

Transformers have been successfully applied in the field of video-based 3D human pose estimation. However, the high computational costs of these video pose transformers (VPTs) make them impractical on resource-constrained devices. In this paper, we present a hierarchical plug-and-play pruning-and-recovering framework, called Hierarchical Hourglass Tokenizer (H2OT), for efficient transformer-based 3D human pose estimation from videos. H2OT begins with progressively pruning pose tokens of redundant frames and ends with recovering full-length sequences, resulting in a few pose tokens in the intermediate transformer blocks and thus improving the model efficiency. It works with two key modules, namely, a Token Pruning Module (TPM) and a Token Recovering Module (TRM). TPM dynamically selects a few representative tokens to eliminate the redundancy of video frames, while TRM restores the detailed spatio-temporal information based on the selected tokens, thereby expanding the network output to the original full-length temporal resolution for fast inference. Our method is general-purpose: it can be easily incorporated into common VPT models on both seq2seq and seq2frame pipelines while effectively accommodating different token pruning and recovery strategies. In addition, our H2OT reveals that maintaining the full pose sequence is unnecessary, and a few pose tokens of representative frames can achieve both high efficiency and estimation accuracy. Extensive experiments on multiple benchmark datasets demonstrate both the effectiveness and efficiency of the proposed method.

Cross-domain few-shot object detection (CD-FSOD) remains a challenging problem for existing object detectors and few-shot learning approaches, particularly when generalizing across distinct domains. As part of NTIRE 2026, we hosted the second CD-FSOD Challenge to systematically evaluate and promote progress in detecting objects in unseen target domains under limited annotation conditions. The challenge received strong community interest, with 128 registered participants and a total of 696 submissions. Among them, 31 teams actively participated, and 19 teams submitted valid final results. Participants explored a wide range of strategies, introducing innovative methods that push the performance frontier under both open-source and closed-source tracks. This report presents a detailed overview of the NTIRE 2026 CD-FSOD Challenge, including a summary of the submitted approaches and an analysis of the final results across all participating teams. Challenge Codes: https://github.com/ohMargin/NTIRE2026_CDFSOD.

Representations on the Symmetric Positive Definite (SPD) manifold have garnered significant attention across different applications. In contrast, the manifold of full-rank correlation matrices, a normalized alternative to SPD matrices, remains largely underexplored. This paper introduces Riemannian networks over the correlation manifold, leveraging five recently developed correlation geometries. We systematically extend basic layers, including Multinomial Logistic Regression (MLR), Fully Connected (FC), and convolutional layers, to these geometries. Besides, we present methods for accurate backpropagation for two correlation geometries. Experiments comparing our approach against existing SPD and Grassmannian networks demonstrate its effectiveness.

Convolutional neural networks (CNNs) have achieved astonishing advances over the past decade, pushing the state-of-the-art in several computer vision tasks. CNNs are capable of learning robust representations of the data directly from RGB pixels. However, most image data is usually available in compressed format, of which the JPEG is the most widely used due to transmission and storage purposes. For this motive, a preliminary decoding process that has a high computational load and memory usage is demanded. Image decoding can be a performance bottleneck for devices with limited computational resources, such as embedded devices, even when hardware accelerators are used. For this reason, deep learning methods capable of learning directly from the compressed domain have been gaining attention in recent years. These methods usually extract a frequency domain representation of the image, like DCT, by a partial decoding, and then make adaptation to typical CNN architectures to work with it. In this paper, we perform an in-depth study of the computational cost of deep models designed for the frequency domain, evaluating the cost of decoding and passing images through the network. We notice that previous work increased the model's computational complexity to accommodate for the compressed images, nullifying the speed up gained by not decoding images. We propose to remove the changes to the model that increase the computational cost, replacing it with our designed lightweight stems. This way, we can take full advantage of the speed-up obtained by avoiding the decoding. Our strategies were successful in generating models that balance efficiency and effectiveness, allowing deep models to be deployed in a wider array of devices. We achieve up to 25.91% reduction in computational complexity (FLOPs), while only decreasing accuracy in up to 2.97%. We also propose the efficiency-effectiveness score SE to highlight models with favorable trade-offs between accuracy, computational cost and number of parameters.

Poisoning input views of 3D reconstruction systems has been recently studied.However, we identify that existing studies simply backpropagate adversarial gradients through the 3D reconstruction pipeline as a whole, without uncovering the new vulnerability rooted in specific modules of the 3D reconstruction pipeline.In this paper, we argue that the structure-from-motion (SfM) initialization, as the geometric core of many widely used reconstruction systems, can be targeted to achieve strong poisoning effects. To this end, we propose PoInit-of-View, which optimizes adversarial perturbations to intentionally introduce cross-view gradient inconsistencies at projections of corresponding 3D points. These inconsistencies disrupt keypoint detection and feature matching, thereby corrupting pose estimation and triangulation within SfM, eventually resulting in low-quality rendered views. We also provide a theoretical analysis that connects cross-view inconsistency to correspondence collapse.Experimental results demonstrate the effectiveness of our \name on diverse 3D reconstruction systems and datasets, surpassing the single-view-based method by 25.1% in PSNR and 16.5% in SSIM in black-box transfer settings, such as 3DGS to NeRF.

Hyperbolic neural networks (HNNs) have shown remarkable success in representing hierarchical and tree-like structures, yet most existing work relies on the Poincaré ball and hyperboloid models. While these models admit closed-form Riemannian operators, their constrained nature potentially leads to numerical instabilities, especially near model boundaries. In this work, we explore the Proper Velocity (PV) manifold, an unconstrained representation of hyperbolic space rooted in Einstein’s special relativity, as a stable alternative. We first establish the complete Riemannian toolkit of the PV space. Building on this foundation, we introduce Proper Velocity Neural Networks (PVNNs) with core layers including Multinomial Logistic Regression (MLR), Fully Connected (FC), convolutional, activation, and batch normalization layers. Extensive experiments across four domains, namely numerical stability, graph node classification, image classification, and genomic sequence learning, demonstrate the stability and effectiveness of PVNNs.

Direct Preference Optimization (DPO) has been proposed as an effective and efficient alternative to reinforcement learning from human feedback (RLHF). However, neither RLHF nor DPO take into account the fact that learning certain preferences is more difficult than learning other preferences, rendering the optimization process suboptimal. To address this gap in text-to-image generation, we recently proposed Curriculum-DPO, a method that organizes image pairs by difficulty. In this paper, we introduce Curriculum-DPO++, an enhanced method that combines the original data-level curriculum with a novel model-level curriculum. More precisely, we propose to dynamically increase the learning capacity of the denoising network as training advances. We implement this capacity increase via two mechanisms. First, we initialize the model with only a subset of the trainable layers used in the original Curriculum-DPO. As training progresses, we sequentially unfreeze layers until the configuration matches the full baseline architecture. Second, as the fine-tuning is based on Low-Rank Adaptation (LoRA), we implement a progressive schedule for the dimension of the low-rank matrices. Instead of maintaining a fixed capacity, we initialize the low-rank matrices with a dimension significantly smaller than that of the baseline. As training proceeds, we incrementally increase their rank, allowing the capacity to grow until it converges to the same rank value as in Curriculum-DPO. Furthermore, we propose an alternative ranking strategy to the one employed by Curriculum-DPO. Finally, we compare Curriculum-DPO++ against Curriculum-DPO and other state-of-the-art preference optimization approaches on nine benchmarks, outperforming the competing methods in terms of text alignment, aesthetics and human preference. Our code is available at https://github.com/CroitoruAlin/Curriculum-DPO.

With advances in imitation learning (IL) and large-scale driving datasets, end-to-end autonomous driving (E2E-AD) has made great progress recently. Currently, IL-based methods have become a mainstream paradigm: models rely on standard driving behaviors given by experts, and learn to minimize the discrepancy between their actions and expert actions. However, this objective of "only driving like the expert" suffers from limited generalization: when encountering rare or unseen long-tail scenarios outside the distribution of expert demonstrations, models tend to produce unsafe decisions in the absence of prior experience. This raises a fundamental question: Can an E2E-AD system make reliable decisions without any expert action supervision? Motivated by this, we propose a unified framework named Risk-aware World Model Predictive Control (RaWMPC) to address this generalization dilemma through robust control, without reliance on expert demonstrations. Practically, RaWMPC leverages a world model to predict the consequences of multiple candidate actions and selects low-risk actions through explicit risk evaluation. To endow the world model with the ability to predict the outcomes of risky driving behaviors, we design a risk-aware interaction strategy that systematically exposes the world model to hazardous behaviors, making catastrophic outcomes predictable and thus avoidable. Furthermore, to generate low-risk candidate actions at test time, we introduce a self-evaluation distillation method to distill riskavoidance capabilities from the well-trained world model into a generative action proposal network without any expert demonstration. Extensive experiments show that RaWMPC outperforms state-of-the-art methods in both in-distribution and out-of-distribution scenarios, while providing superior decision interpretability.

Line coverage of urban-scale routes by multiple vehicles can be formulated as the Capacitated Arc Routing Problem (CARP). Early solutions, including constructive heuristics, meta-heuristics, and exact algorithms, struggle to achieve efficient runtimes. Recently, neural network (NN)-based methods have emerged as promising alternatives for efficiently solving CARP; however, their solution quality significantly lags behind non-NN approaches. This significant gap mainly results from the inappropriate modeling and learning of non-Euclidean graphs, traversal direction and capacity constraints. In this paper, we introduce an NN-based solver tailored for these complexities, which significantly narrows the gap with advanced meta-heuristics while achieving much shorter runtimes. First, we propose the direction-aware attention model (DaAM) to incorporate directionality into the embedding process, facilitating more effective one-stage decision-making. Second, we design a supervised reinforcement learning scheme that involves supervised pre-training to establish a robust initial policy for subsequent reinforcement fine-tuning. It proves particularly valuable for solving CARP that has a higher complexity than the node routing problems (NRPs). Finally, a path optimization method is introduced to adjust the depot return positions within the path generated by DaAM. Experiments show that DaAM surpasses constructive heuristics and achieves decision quality comparable to state-of-the-art meta-heuristics for the first time while maintaining superior efficiency, even in large-scale CARP instances. The code and datasets are provided in the link and will be released on GitHub.

In federated learning, Transformer, as a popular architecture, faces critical challenges in defending against gradient attacks and improving model performance in both Computer Vision (CV) and Natural Language Processing (NLP) tasks. It has been revealed that the gradient of Position Embeddings (PEs) in Transformer contains sufficient information, which can be used to reconstruct the input data. To mitigate this issue, we introduce a Masked Jigsaw Puzzle (MJP) framework. MJP starts with random token shuffling to break the token order, and then a learnable unknown (unk) position embedding is used to mask out the PEs of the shuffled tokens. In this manner, the local spatial information which is encoded in the position embeddings is disrupted, and the models are forced to learn feature representations that are less reliant on the local spatial information. Notably, with the careful use of MJP, we can not only improve models' robustness against gradient attacks, but also boost their performance in both vision and text application scenarios, such as classification for images (e.g., ImageNet-1 K) and sentiment analysis for text (e.g., Yelp and Amazon). Experimental results suggest that MJP is a unified framework for different Transformer-based models in both vision and language tasks.

Multi-view clustering (MVC) aims to uncover the latent structure of multi-view data by learning view-common and view-specific information. Although recent studies have explored hyperbolic representations for better tackling the representation gap between different views, they focus primarily on instance-level alignment and neglect global semantic consistency, rendering them vulnerable to view-specific information (e.g., noise and cross-view discrepancies). To this end, this paper proposes a novel Wasserstein-Aligned Hyperbolic (WAH) framework for multi-view clustering. Specifically, our method exploits a view-specific hyperbolic encoder for each view to embed features into the Lorentz manifold for hierarchical semantic modeling. Whereafter, a global semantic loss based on the hyperbolic sliced-Wasserstein distance is introduced to align manifold distributions across views. This is followed by soft cluster assignments to encourage cross-view semantic consistency. Extensive experiments on multiple benchmarking datasets show that our method can achieve SOTA clustering performance.

Artistic style transfer (AST) is a form of image editing that generates a stylized output, maintaining the original image's content structure while applying a learned artistic style. Current methods for achieving this can be broadly categorized into small-scale model-based methods (SMM) and large-scale pre-trained model-based methods (LMM). The former often results in stylized images with obvious artifacts and disharmonious patterns, while the latter typically struggle to preserve the content structure of the input image. To address these challenges, we propose a step-aware artistic style transfer method, called SAAST, which learns content and style across denoising steps and produces highly realistic stylized images. Specifically, we observe that during the denoising process, the low-noise stages generate local style patterns, while the high-noise stages are more likely to influence the content structure. Thus, we first propose a Step-aware Style Representation Module (SSRM) that effectively captures style patterns from artistic images while accounting for the varying controllable conditions required at different denoising steps in large-scale pre-trained models. Secondly, we propose a Step-aware Content Reconstruction Module (SCRM), which can capture the content features of the input content image and guide the large-scale pre-trained model to maintain the input image's content structures during the style transfer process. Finally, the extensive experiments show that SAAST significantly surpasses existing SOTA methods.

Human-centric intelligence is evolving in the foundation-model era, with growing emphasis on scale, transferability, and general-purpose modeling. Yet it has not fully integrated with foundation models to achieve the comparable progress seen in them. More importantly, recent advances across this broad landscape remain fragmented across tasks, modalities, and research communities, leaving their intrinsic conceptual and methodological connections unclear. To bridge these divides and rethink human-centric intelligence in the foundation-model era, we introduce a full-spectrum human context taxonomy that integrates six interconnected levels by viewing humans as observable subjects through visual appearance and spatial geometry, as dynamic actors through kinematic dynamics and interaction modeling, and as situated agents through world simulation and embodied agency. We next present the methodological foundations of the field, covering human-centric data families, computational architecture paradigms, and representative training and inference optimization strategies. We then systematically review representative methods across these levels and organize the associated datasets, benchmarks, and evaluation metrics. We further discuss open challenges and promising research directions toward human-centric intelligence that is scalable, trustworthy, physically grounded, and deployable, aiming to provide a coherent framework and practical reference for advancing the field. Finally, we provide a systematically organized and continuously updated collection of human-centric AI literature and resources on our project page.

Visuomotor policies aim to learn complex manipulation tasks from expert demonstrations. However, generating smooth and coherent trajectories remains challenging, as it requires balancing proximal precision with distal foresight. Existing approaches typically focus on optimizing intra-chunk action distributions, often neglecting the inter-chunk coherence. Consequently, inter-chunk discontinuities significantly impede the learning of coherent long-horizon actions. To overcome this limitation and achieve a synergetic balance between precision and foresight, we propose FocalPolicy, a foresight-aware visuomotor policy that combines Frequency-Optimized Chunking with Locally Anchored flow matching. We introduce a foresight composite objective that supervises time-domain alignment within the proximal actions while regularizing frequency-domain structure over multiple future action chunks to improve cross-chunk coherence. To efficiently learn complex action distributions, we design locally anchored campling to enhance target signal propagation efficiency during consistency flow matching training. Extensive experiments demonstrate that our method consistently outperforms existing approaches while further validating the effective generalizability of our proposed modules to other baseline models. The project will be released as open source.

Existing text-to-motion (T2M) generation models typically rely on pretrained large language models to encode textual inputs. However, these models, trained on generic text corpora, lack explicit alignment between motion-related words (e.g., "clockwise'', "quickly'') and human skeletal movements. This misalignment, fundamentally rooted in the word embedding layers, severely limits the ability of T2M models to understand and generalize fine-grained motion semantics. To tackle this issue, we propose Motion-Aligned Text Encoding (MATE), a novel framework that explicitly incorporates motion semantics into the word embedding layers of large language models to enhance text-motion alignment for motion generation. To address the challenge of inherent semantic entanglement in motion sequences, MATE introduces two key components: 1) a motion localization strategy that establishes localized correspondences between sub-texts and motion segments, enabling soft attention guidance for semantic localization; and 2) a motion disentanglement module that isolates word-specific motion semantics via contrastive kinematic prototypes, ensuring word-level alignment between linguistic and kinematic representations. Remarkably, language models enhanced with MATE can be seamlessly integrated into existing T2M methods, significantly surpassing state-of-the-art performance on two standard benchmarks with minimal modifications. Codes and pretrained models will be released upon acceptance.

While 3DGS has emerged as a high-fidelity scene representation, encoding rich, general-purpose features directly from its primitives remains under-explored. We address this gap by introducing Chorus, a multi-teacher pretraining framework that learns a holistic feed-forward 3D Gaussian Splatting (3DGS) scene encoder by distilling complementary signals from 2D foundation models. Chorus employs a shared 3D encoder and teacher-specific projectors to learn from language-aligned, generalist, and object-aware teachers, encouraging a shared embedding space that captures signals from high-level semantics to fine-grained structure. We evaluate Chorus on a wide range of tasks: open-vocabulary semantic and instance segmentation, linear and decoder probing, as well as data-efficient supervision. Besides 3DGS, we also test Chorus on several benchmarks that only support point clouds by pretraining a variant using only Gaussians' centers, colors, estimated normals as inputs. Interestingly, this encoder shows strong transfer and outperforms the point clouds baseline while using 39.9 times fewer training scenes. Finally, we propose a render-and-distill adaptation that facilitates out-of-domain finetuning. Our code and model will be released upon publication.

Traditional video compression methods perform well at high bitrates but struggle to preserve fine-grained semantic information at low bitrates. Recently, with the blossoming of Multimodal Large Language Models (MLLMs), Cross-modal compression techniques offer prospective solutions for improving video compression under low-bitrate conditions. In this paper, we propose a unified Cross-Modality Video Compression (CMVC) framework that integrates multimodal representations and video generative models. The encoder disentangles video into spatial and temporal components, which are mapped to compact cross modal representations using MLLMs. During decoding, different encoding-decoding modes are employed to acquire various video reconstruction qualities, including Text-Text-to-Video (TT2V) for semantic preservation and Image-Text-to-Video (IT2V) for perceptual consistency. Additionally, we elaborate on an efficient frame interpolation model using Low-Rank Adaptation (LoRA) to improve the perceptual quality. Experimental results demon strate that TT2V achieves effective semantic reconstruction, while IT2V ensures competitive perceptual consistency. These findings suggest the potential of leveraging multimodal priors to improve video compression, offering promising future research directions.