CVPR 2025 Papers

We introduce HunyuanPortrait, a diffusion-based condition control method that employs implicit representations for highly controllable and lifelike portrait animation. Given a single portrait image as an appearance reference and video clips as driving templates, HunyuanPortrait can animate the character in the reference image by the facial expression and head pose of the driving videos. In our framework, we utilize pre-trained encoders to achieve the decoupling of portrait motion information and identity in videos. To do so, implicit representation is adopted to encode motion information and is employed as control signals in the animation phase. By leveraging the power of stable video diffusion as the main building block, we carefully design adapter layers to inject control signals into the denoising unet through attention mechanisms. These bring spatial richness of details and temporal consistency. HunyuanPortrait also exhibits strong generalization performance, which can effectively disentangle appearance and motion under different image styles. Our framework outperforms existing methods, demonstrating superior temporal consistency and controllability. Our project is available at https://kkakkkka.github.io/HunyuanPortrait.

Despite remarkable advancements in text-to-image person re-identification (TIReID) facilitated by the breakthrough of cross-modal embedding models, existing methods often struggle to distinguish challenging candidate images due to intrinsic limitations, such as network architecture and data quality. To address these issues, we propose an Interactive Cross-modal Learning framework (ICL), which leverages human-centered interaction to enhance the discriminability of text queries through external multimodal knowledge. To achieve this, we propose a plug-and-play Test-time Humane-centered Interaction (TUI) module, which performs visual question answering focused on human characteristics, facilitating multi-round interactions with a multimodal large language model (MLLM) to align query intent with latent target images. Specifically, TUI refines user queries based on the MLLM responses to reduce the gap to the best-matching images, thereby boosting ranking accuracy. Additionally, to address the limitation of low-quality training texts, we introduce a novel Reorganization Data Augmentation (RDA) strategy based on information enrichment and diversity enhancement to enhance query discriminability by enriching, decomposing, and reorganizing person descriptions. Extensive experiments on four TIReID benchmarks, i.e., CUHK-PEDES, CFG-PEDES RSTPReid, RSTPReid, and UFine6926, demonstrate that our method achieves remarkable performance with substantial improvement. The code will be released publicly.

Model merging combines expert models---each finetuned from a shared foundation model on diverse tasks and domains---into a single, more capable base model. However, existing model merging approaches assume all experts to be available simultaneously. In reality, new tasks and domains emerge continuously, prompting the need for a dynamic process of integrating these experts over time, which we call temporal model merging. The temporal dimension introduces unique challenges not addressed in prior work: At each task, should expert training start from merged previous experts or the original base model? Should all models be merged at every time step? Which merging techniques are best suited for temporal merging? Should different strategies be used for the training initialization and deployment phases? To tackle these questions, we propose a unified framework called TIME---Temporal Integration of Model Expertise---that defines temporal model merging across three axes: (1) Initialization Phase, (2) Deployment Phase, and (3) Merging Technique. Utilizing TIME, we study temporal model merging across model sizes, tasks, and compute budgets on the large-scale FoMo-in-Flux benchmark for continual multimodal pretraining. Systematic experiments across TIME and FoMo-in-Flux allow us to arrive at several crucial key insights for temporal model merging to better understand current limits and best practices for successful model merging across time.

How can we enable models to comprehend video anomalies occurring over varying temporal scales and contexts?Traditional Video Anomaly Understanding (VAU) methods focus on frame-level anomaly prediction, often missing the interpretability of complex and diverse real-world anomalies. Recent multimodal approaches leverage visual and textual data but lack hierarchical annotations that capture both short-term and long-term anomalies.To address this challenge, we introduce HIVAU-70k, a large-scale benchmark for hierarchical video anomaly understanding across any granularity. We develop a semi-automated annotation engine that efficiently scales high-quality annotations by combining manual video segmentation with recursive free-text annotation using large language models (LLMs). This results in over 70,000 multi-granular annotations organized at clip-level, event-level, and video-level segments.For efficient anomaly detection in long videos, we propose the Anomaly-focused Temporal Sampler (ATS). ATS integrates an anomaly scorer with a density-aware sampler to adaptively select frames based on anomaly scores, ensuring that the multimodal LLM concentrates on anomaly-rich regions, which significantly enhances both efficiency and accuracy.Extensive experiments demonstrate that our hierarchical instruction data markedly improves anomaly comprehension. The integrated ATS and visual-language model outperform traditional methods in processing long videos.Our benchmark and model will be publicly available.

In this work we address various shape matching problems that can be cast as finding cyclic paths in a product graph. This involves for example 2D-3D shape matching, 3D shape matching, or the matching of a contour to a graph. In this context, matchings are typically obtained as the minimum cost cycle in the product graph. Instead, inspired by related works on model-based image segmentation (Schoenemann and Cremers 2009), we consider minimum ratio cycles, which we combine with the recently introduced conjugate product graph in order to allow for higher-order matching costs. With that, on the one hand we avoid the bias of obtaining matchings that involve fewer/shorter edges, while on the other hand we are able to impose powerful geometric regularisation, e.g. to avoid zig-zagging. In our experiments we demonstrate that this not only leads to improved matching accuracy in most cases, but also to significantly reduced runtimes (up to two orders of magnitude, depending on the setting). Our GPU implementations are publicly available: https://github.com/paul0noah/product-graph-cycles/.

Semantic segmentation from RGB cameras is essential to the perception of autonomous flying vehicles. The stability of predictions through the captured videos is paramount to their reliability and, by extension, to the trustworthiness of the agents. In this paper, we propose a lightweight video semantic segmentation approach--suited to onboard real-time inference--achieving high temporal consistency on aerial data through Semantic Similarity Propagation across frames. SSP temporally propagates the predictions of an efficient image segmentation model with global registration alignment to compensate for camera movements. It combines the current estimation and the prior prediction with linear interpolation using weights computed from the features similarities of the two frames. Because data availability is a challenge in this domain, we propose a consistency-aware Knowledge Distillation training procedure for sparsely labeled datasets with few annotations. Using a large image segmentation model as a teacher to train the efficient SSP, we leverage the strong correlations between labeled and unlabeled frames in the same training videos to obtain high-quality supervision on all frames. KD-SSP obtains a significant temporal consistency increase over the base image segmentation model of 12.5% and 6.7% TC on UAVid and RuralScapes respectively, with higher accuracy and comparable inference speed. On these aerial datasets, KD-SSP provides a superior segmentation quality and inference speed trade-off than other video methods proposed for general applications and shows considerably higher consistency. The code will be made publicly available upon acceptance.

Test-time adaptation using vision-language models (such as CLIP) to quickly adjust to distributional shifts of downstream tasks has shown great potential. Despite significant progress, existing methods are still limited to single-task test-time adaptation scenarios and have not effectively explored the issue of multi-task adaptation. To address this practical problem, we propose a novel Hierarchical Knowledge Prompt Tuning (HKPT) method, which achieves joint adaptation to multiple target domains by mining more comprehensive source domain discriminative knowledge and hierarchically modeling task-specific and task-shared knowledge. Specifically, HKPT constructs a CLIP prompt distillation framework that utilizes the broader source domain knowledge of large teacher CLIP to guide prompt tuning for lightweight student CLIP from multiple views during testing. Meanwhile, HKPT establishes task-specific dual dynamic knowledge graph to capture fine-grained contextual knowledge from continuous test data. To fully exploit the complementarity among multiple target tasks, HKPT employs an adaptive task grouping strategy for achieving inter-task knowledge sharing. Furthermore, HKPT can seamlessly transfer to basic single-task test-time adaptation scenarios while maintaining robust performance. Extensive experimental results in both multi-task and single-task test-time adaptation settings demonstrate that our HKPT significantly outperforms state-of-the-art methods.

Despite significant advances in the field of specular highlight removal in recent years, existing methods predominantly focus on natural images, where highlights typically appear on raised or edged surfaces of objects. These highlights are often small and sparsely distributed. However, for text images such as cards and posters, the flat surfaces reflect light uniformly, resulting in large areas of highlights. Current methods struggle with these large-area highlights in text images, often producing severe visual artifacts or noticeable discrepancies between filled pixels and the original image in the central high-intensity highlight areas. To address these challenges, we propose the Hierarchical Adaptive Filtering Network (HAFNet). Our approach performs filtering at both the downsampled deep feature layer and the upsampled image reconstruction layer. By designing and applying the Adaptive Comprehensive Filtering Module (ACFM) and Adaptive Dilated Filtering Module (ADFM) at different layers, our method effectively restores semantic information in large-area specular highlight regions and recovers detail loss at various scales. The required filtering kernels are pre-generated by a prediction network, allowing them to adaptively adjust according to different images and their semantic content, enabling robust performance across diverse scenarios. Additionally, we utilize Unity3D to construct a comprehensive large-area highlight dataset featuring images with rich texts and complex textures. Experimental results on various datasets demonstrate that our method outperforms state-of-the-art approaches.

Despite advancements in multimodal large language models (MLLMs), current approaches struggle in medium-to-long video understanding due to frame and context length limitations. As a result, these models often depend on frame sampling, which risks missing key information over time and lacks task-specific relevance. To address these challenges, we introduce **HierarQ**, a task-aware hierarchical Q-Former based framework that sequentially processes frames to bypass the need for frame sampling, while avoiding LLM's context length limitations. We introduce a lightweight two-stream language-guided feature modulator to incorporate task awareness in video understanding, with the entity stream capturing frame-level object information within a short context and the scene stream identifying their broader interactions over longer period of time. Each stream is supported by dedicated memory banks which enables our proposed **Hierar**chical **Q**uerying transformer (HierarQ) to effectively capture short and long-term context. Extensive evaluations on **10** video benchmarks across video understanding, question answering, and captioning tasks demonstrate HierarQ's state-of-the-art performance across most datasets, proving its robustness and efficiency for comprehensive video analysis. All code will be made available upon acceptance.

High-resolution image inputs allow Large Vision-Language Models (LVLMs) to capture finer visual details, improving comprehension. However, the increased training and computational costs associated with such inputs pose significant challenges. A common approach to mitigate these costs involves slicing the input into uniform patches using sliding windows, each aligned with the vision encoder's input size. While efficient, this method fragments the input, disrupting the continuity of context, which negatively impacts cross-patch perception tasks. To address these limitations, we propose HiRes-LLaVA, a novel framework designed to efficiently process high-resolution inputs of any size without altering the original contextual and geometric information. HiRes-LLaVA introduces two key components: (i) a SliceRestore Adapter (SRA) that reconstructs sliced patches into their original form, enabling efficient extraction of both global and local features through down-up-sampling and convolutional layers, and (ii) a Self-Mining Sampler (SMS) that compresses visual tokens based on internal relationships, preserving original context and positional information while reducing training overhead. To assess the ability of handling context fragmentation, we construct a new benchmark, EntityGrid-QA, consisting of edge-related tasks. Extensive experiments demonstrate the superiority of HiRes-LLaVA on both existing public benchmarks and EntityGrid-QA. For example, with SRA, our method achieves a performance improvement of 12% over state-of-the-art LVLMs in addressing fragmentation issues. Additionally, our SMS outperforms other visual token downsamplers, while offering high data efficiency.

Currently, the demand for higher video quality has grown significantly. However, satellite video has low resolution, complex motion, and weak textures. Haze interference further exacerbates the loss of motion information and texture details, hindering effective spatiotemporal feature fusion and fine-grained feature mining. This presents significant challenges for subsequent super-resolution (SR) reconstruction, especially at continuous scales. To address these problems, this paper models the double-degradation process of hazy low-quality satellite videos and proposes a novel network to learn the optimal joint degradation pattern (ODPNet) for continuous-scale SR of hazy satellite videos. First, we design a prior-based feature soft dehazing module to eliminate haze interference at the feature level. Second, we develop a spatiotemporal self-attention (SSA) to capture long-range feature dependencies, thereby achieving effective spatiotemporal feature fusion. Third, we devise a tri-branch cross-aggregation block (TCB) to enhance feature representations of weak textures in satellite videos by effectively aggregating contextual information. Finally, we propose a cross-scale feature Top-k selection Transformer (CFTST), which aims to adaptively select and aggregate cross-scale latent codes to learn feature representations of satellite videos at arbitrary resolutions, thus enabling continuous-scale SR. Experiments show that ODPNet outperforms existing methods and achieves a better balance between model parameters and performance.

Recent contrastive multimodal vision-language models like CLIP have demonstrated robust open-world semantic understanding, becoming the standard image backbones for vision-language applications. However, recent findings suggest high semantic similarity between well-trained unimodal encoders, which raises a key question: Are semantically similar embedding spaces separated only by simple projection transformations? To validate this, we propose a novel framework that aligns vision and language using frozen unimodal encoders. It involves selecting semantically similar encoders in the latent space, curating a concept-rich dataset of image-caption pairs, and training simple MLP projectors. We evaluated our approach on various tasks involving both strong unimodal vision (0-shot localization) and language encoders (multi-lingual, long context) and show that simple Projectors retain unimodal capabilities in joint embedding space. Furthermore, our best model, utilizing DINOv2 and All-Roberta-Large text encoder, achieves 76(%) accuracy on ImageNet with a 20-fold reduction in data and 65-fold reduction in compute requirements compared to multimodal alignment where models are trained from scratch. The proposed framework enhances the accessibility of multimodal model development while enabling flexible adaptation across diverse scenarios. Code and curated datasets are available at github.com/mayug/freeze-align.

Existing methodologies for animating portrait images face significant challenges, particularly in handling non-frontal perspectives, rendering dynamic objects around the portrait, and generating immersive, realistic backgrounds. In this paper, we introduce the first application of a pretrained transformer-based video generative model that demonstrates strong generalization capabilities and generates highly dynamic, realistic videos for portrait animation, effectively addressing these challenges. The adoption of a new video backbone model makes previous U-Net-based methods for identity maintenance, audio conditioning, and video extrapolation inapplicable. To address this limitation, we design an identity reference network consisting of a causal 3D VAE combined with a stacked series of transformer layers, ensuring consistent facial identity across video sequences. Additionally, we investigate various speech audio conditioning and motion frame mechanisms to enable the generation of continuous video driven by speech audio. Our method is validated through experiments on benchmark and newly proposed wild datasets, demonstrating substantial improvements over prior methods in generating realistic portraits characterized by diverse orientations within dynamic and immersive scenes.

Hallucinations pose a significant challenge to the reliability of large vision-language models, making their detection essential for ensuring accuracy in critical applications. Current detection methods often rely on computationally intensive models, leading to high latency and resource demands. Their definitive outcomes also fail to account for real-world scenarios where the line between hallucinated and truthful information is unclear. To address these issues, we propose HalLoc, a dataset designed for efficient, probabilistic hallucination detection. It features 150K token-level annotated samples, including hallucination types, across Visual Question Answering (VQA), instruction-following, and image captioning tasks. This dataset facilitates the development of models that detect hallucinations with graded confidence, enabling more informed user interactions. Additionally, we introduce a baseline model trained on HalLoc, offering low-overhead, concurrent hallucination detection during generation. The model can be seamlessly integrated into existing VLMs, improving reliability while preserving efficiency. The prospect of a robust plug-and-play hallucination detection module opens new avenues for enhancing the trustworthiness of vision-language models in real-world applications. The HalLoc dataset and code are publicly available at: https://github.com/dbsltm/cvpr25_halloc.

Low-Light Image Enhancement (LLIE) is a crucial computer vision task that aims to restore detailed visual information from corrupted low-light images. Many existing LLIE methods are based on standard RGB (sRGB) space, which often produce color bias and brightness artifacts due to inherent high color sensitivity in sRGB. While converting the images using Hue, Saturation and Value (HSV) color space helps resolve the brightness issue, it introduces significant red and black noise artifacts. To address this issue, we propose a new color space for LLIE, namely Horizontal/Vertical-Intensity (HVI), defined by polarized HS maps and learnable intensity. The former enforces small distances for red coordinates to remove the red artifacts, while the latter compresses the low-light regions to remove the black artifacts. To fully leverage the chromatic and intensity information, a novel Color and Intensity Decoupling Network (CIDNet) is further introduced to learn accurate photometric mapping function under different lighting conditions in the HVI space. Comprehensive results from benchmark and ablation experiments show that the proposed HVI color space with CIDNet outperforms the state-of-the-art methods on 10 datasets. The code is available at https://github.com/Fediory/HVI-CIDNet.

Deep Unfolding Networks (DUNs) have risen to prominence due to their interpretability and superior performance for image Compressive Sensing (CS). However, existing DUNs still face significant issues, such as the insufficient representation capability of single-scale image information during the iterative reconstruction phase and loss of feature information, which fundamentally limit the further enhancement of image CS performance. In this paper, we propose Homotopy Unfolding Network (HUNet) for image CS, which enables phase-by-phase reconstruction of images along homotopy path. Specifically, each iteration step of the traditional homotopy algorithm is mapped to a Multi-scale Homotopy Iterative Module (MHIM), which includes U-shaped stacked window-based Transformer blocks capable of efficient feature extraction. Within the MHIM, we design the deep homotopy continuation strategy to ensure the interpretability of the homotopy algorithm and facilitate feature learning. Additionally, we introduce a dual-path feature fusion module to mitigate the loss of high-dimensional feature information during the transmission between iterative phases, thereby maximizing the preservation of details in the reconstructed image. Extensive experiments indicate that HUNet achieves superior image reconstruction results compared to existing state-of-the-art methods. The source code is available at https://github.com/ICSResearch/HUNet.

Co-speech gestures are crucial non-verbal cues that enhance speech clarity and expressiveness in human communication, which have attracted increasing attention in multimodal research. While the existing methods have made strides in gesture accuracy, challenges remain in generating diverse and coherent gestures, as most approaches assume independence among multimodal inputs and lack explicit modeling of their interactions. In this work, we propose a novel multimodal learning method named HOP for co-speech gesture generation that captures the heterogeneous entanglement between gesture motion, audio rhythm, and text semantics, enabling the generation of coordinated gestures. By leveraging spatiotemporal graph modeling, we achieve the alignment of audio and action. Moreover, to enhance modality coherence, we build the audio-text semantic representation based on a reprogramming module, which is beneficial for cross-modality adaptation. Our approach enables the trimodal system to learn each other's features and represent them in the form of topological entanglement. Extensive experiments demonstrate that HOP achieves state-of-the-art performance, offering more natural and expressive co-speech gesture generation. More information, codes, and demos are available here: https://star-uu-wang.github.io/HOP/.

Text-to-video (T2V) generation has made tremendous progress in generating complicated scenes based on texts. However, human-object interaction (HOI) often cannot be precisely generated by current T2V models due to the lack of large-scale videos with accurate captions for HOI. To address this issue, we introduce HOIGen-1M, the first large-scale dataset for HOI Generation, consisting of over one million high-quality videos collected from diverse sources. In particular, to guarantee the high quality of videos, we first design an efficient framework to automatically curate HOI videos using the powerful multimodal large language models (MLLMs), and then the videos are further cleaned by human annotators. Moreover, to obtain accurate textual captions for HOI videos, we design a novel video description method based on a Mixture-of-Multimodal-Experts (MoME) strategy that not only generates expressive captions but also eliminates the hallucination by individual MLLM. Furthermore, due to the lack of an evaluation framework for generated HOI videos, we propose two new metrics to assess the quality of generated videos in a coarse-to-fine manner. Extensive experiments reveal that current T2V models struggle to generate high-quality HOI videos and confirm that our HOIGen-1M dataset is instrumental for improving HOI video generation.