CVPR 2025 Papers

We introduce SynthLight, a diffusion model for portrait relighting. We frame image relighting as a re-rendering problem, where pixels are transformed in response to changes in environmental lighting. Using a physically-based rendering engine, we create a dataset to simulate this lighting-conditioned transformation with 3D head assets under varying lighting. We propose training and inference strategies to bridge the gap between the synthetic and real image domains: (1) multi-task training leveraging real human portraits without lighting labels; (2) an inference time diffusion sampling scheme based on classifier-free guidance leveraging the input portrait to better preserve details. Our method generalizes to diverse real portraits and produces realistic illumination effects, including specular highlights and cast shadows, while preserving identity. Our quantitative experiments on light stage data demonstrate results comparable to state-of-the-art relighting methods. Our qualitative results on in-the-wild images showcase rich and unprecedented illumination effects.

Talking head video generation aims to generate a realistic talking head video that preserves the person's identity from a source image and the motion from a driving video. Despite the promising progress made in the field, it remains a challenging and critical problem to generate videos with accurate poses and fine-grained facial details simultaneously. Essentially, facial motion is often highly complex to model precisely, and the one-shot source face image cannot provide sufficient appearance guidance during generation due to dynamic pose changes. To tackle the problem, we propose to jointly learn motion and appearance codebooks and perform multi-scale codebook compensation to effectively refine both the facial motion conditions and appearance features for talking face image decoding. Specifically, the designed multi-scale motion and appearance codebooks are learned simultaneously in a unified framework to store representative global facial motion flow and appearance patterns. Then, we present a novel multi-scale motion and appearance compensation module, which utilizes a transformer-based codebook retrieval strategy to query complementary information from the two codebooks for joint motion and appearance compensation. The entire process produces motion flows of greater flexibility and appearance features with fewer distortions across different scales, resulting in a high-quality talking head video generation framework. Extensive experiments on various benchmarks validate the effectiveness of our approach and demonstrate superior generation results from both qualitative and quantitative perspectives when compared to state-of-the-art competitors. The project page is available at https://shaelynz.github.io/synergize-motion-appearance/.

The remarkable success of Large Language Models (LLMs) has extended to the multimodal domain, achieving outstanding performance in image understanding and generation. Recent efforts to develop unified Multimodal Large Language Models (MLLMs) that integrate these capabilities have shown promising results. However, existing approaches often involve complex designs in model architecture or training pipeline, increasing the difficulty of model training and scaling. In this paper, we propose SynerGen-VL, a simple yet powerful encoder-free MLLM capable of both image understanding and generation. To address challenges identified in existing encoder-free unified MLLMs, we introduce the token folding mechanism and the vision-expert-based progressive alignment pretraining strategy, effectively supporting high-resolution image understanding while reducing training complexity. After being trained on large-scale mixed image-text data with a unified next-token prediction objective, SynerGen-VL achieves or surpasses the performance of existing encoder-free unified MLLMs with comparable or smaller parameter sizes, and narrows the gap with task-specific state-of-the-art models, highlighting a promising path toward future unified MLLMs. Our code and models shall be released.

Video-to-audio generation is essential for synthesizing realistic audio tracks that synchronize effectively with silent videos.Following the perspective of extracting essential signals from videos that can precisely control the mature text-to-audio generative diffusion models, this paper presents how to balance the representation of mel-spectrograms in terms of completeness and complexity through a new approach called Mel Quantization-Continuum Decomposition (Mel-QCD).We decompose the mel-spectrogram into three distinct types of signals, employing quantization or continuity to them, we can effectively predict them from video by a devised video-to-all (V2X) predictor.Then, the predicted signals are recomposed and fed into a ControlNet, along with a textual inversion design, to control the audio generation process.Our proposed Mel-QCD method demonstrates state-of-the-art performance across eight metrics, evaluating dimensions such as quality, synchronization, and semantic consistency.

Predicting future video frames is essential for decision-making systems, yet RGB frames alone often lack the information needed to fully capture the underlying complexities of the real world. To address this limitation, we propose a multi-modal framework for Synchronous Video Prediction (SyncVP) that incorporates complementary data modalities, enhancing the richness and accuracy of future predictions. SyncVP builds on pre-trained modality-specific diffusion models and introduces an efficient spatio-temporal cross-attention module to enable effective information sharing across modalities. We evaluate SyncVP on standard benchmark datasets, such as Cityscapes and BAIR, using depth as an additional modality. We furthermore demonstrate its generalization to other modalities on SYNTHIA with semantic information and ERA5-Land with climate data. Notably, SyncVP achieves state-of-the-art performance, even in scenarios where only one modality is present, demonstrating its robustness and potential for a wide range of applications.

There have been many attempts to leverage multiple diffusion models for collaborative generation, extending beyond the original domain. A prominent approach involves synchronizing multiple diffusion trajectories by mixing the estimated scores to artificially correlate the generation processes. However, existing methods rely on naive heuristics, such as averaging, without considering task specificity. These approaches do not clarify why such methods work and often fail when a heuristic suitable for one task is blindly applied to others. In this paper, we present a probabilistic framework for analyzing why diffusion synchronization works and reveal where heuristics should be focused--modeling correlations between multiple trajectories and adapting them to each specific task. We further identify optimal correlation models per task, achieving better results than previous approaches that apply a single heuristic across all tasks without justification.

Due to the limited scale of multimodal table understanding (MTU) data, model performance is constrained. A straightforward approach is to use multimodal large language models to obtain more samples, but this may cause hallucinations, generate incorrect sample pairs, and cost significantly.To address the above issues, we design a simple yet effective synthesis framework that consists of two independent steps: table image rendering and table question and answer (Q&A) pairs generation.We use table codes (HTML, LaTeX, Markdown) to synthesize images and generate Q&A pairs with large language model (LLM).This approach leverages LLM's high concurrency and low cost to boost annotation efficiency and reduce expenses. By inputting code instead of images, LLMs can directly access the content and structure of the table, reducing hallucinations in table understanding and improving the accuracy of generated Q&A pairs. Finally, we synthesize a large-scale MTU dataset, SynTab, containing 636K images and 1.8M samples costing within \200 in US dollars. We further introduce a generalist tabular multimodal model, SynTab-LLaVA. This model not only effectively extracts local textual content within the table but also enables global modeling of relationships between cells.SynTab-LLaVA achieves SOTA performance on 21 out of 24 in-domain and out-of-domain benchmarks, demonstrating the effectiveness and generalization of our method. The Code is available at \href https://github.com/bang123-box/SynTab-LLaVA SynTab-LLaVA .

We propose a symbolic generative task description language and a corresponding inference engine that can represent arbitrary multimodal tasks as structured symbolic flows. Unlike conventional generative models, which rely on large-scale training and implicit neural representations to learn cross-modal mappings--often with high computational costs and limited flexibility--our framework introduces an explicit symbolic representation composed of three core primitives: functions, parameters, and topological logic. Using a pre-trained language model, our inference engine maps natural language instructions directly to symbolic workflows in a training-free manner. Our framework successfully performs over 12 diverse multimodal generative tasks, demonstrating strong performance and flexibility without requiring task-specific tuning. Experiments show that our method not only matches or outperforms existing state-of-the-art unified models in content quality but also offers greater efficiency, editability, and interruptibility. We believe symbolic task representations provide a cost-effective and extensible foundation for advancing the capabilities of generative AI.

As language models continue to scale, Large Language Models (LLMs) have exhibited emerging capabilities in In-Context Learning (ICL), enabling them to solve language tasks by prefixing a few in-context demonstrations (ICDs) as context. Inspired by these advancements, researchers have extended these techniques to develop Large Multimodal Models (LMMs) with ICL capabilities. However, existing LMMs face a critical issue: they often fail to effectively leverage the visual context in multimodal demonstrations and instead simply follow textual patterns. This indicates that LMMs do not achieve effective alignment between multimodal demonstrations and model outputs. To address this problem, we propose Symbol Demonstration Direct Preference Optimization (SymDPO). Specifically, SymDPO aims to break the traditional paradigm of constructing multimodal demonstrations by using random symbols to replace text answers within instances. This forces the model to carefully understand the demonstration images and establish a relationship between the images and the symbols to answer questions correctly. We validate the effectiveness of this method on multiple benchmarks, demonstrating that with SymDPO, LMMs can more effectively understand the multimodal context within examples and utilize this knowledge to answer questions better. Code is available at https://github.com/APiaoG/SymDPO.

Recent advances in text-guided image editing enable users to perform image edits through simple text inputs, leveraging the extensive priors of multi-step diffusion-based text-to-image models. However, these methods often fall short of the speed demands required for real-world and on-device applications due to the costly multi-step inversion and sampling process involved. In response to this, we introduce SwiftEdit, a simple yet highly efficient editing tool that achieve instant text-guided image editing in 0.23s. The advancement of SwiftEdit lies in its two novel contributions: a one-step inversion framework that enables one-step image reconstruction via inversion and a mask-guided editing technique with our proposed attention rescaling mechanism to perform localized image editing. Extensive experiments are provided to demonstrate the effectiveness and efficiency of SwiftEdit. In particular, SwiftEdit enables instant text-guided image editing, which is extremely faster than previous multi-step methods (at least 50 times faster) while maintain a competitive performance in editing results. Our project is at https://swift-edit.github.io/.

Point cloud (PC) processing tasks--such as completion, upsampling, denoising, and colorization--are crucial in applications like autonomous driving and 3D reconstruction. Despite substantial advancements, prior approaches often address each of these tasks independently, with separate models focused on individual issues. However, this isolated approach fails to account for the fact that defects like incompleteness, low resolution, noise, and lack of color frequently coexist, with each defect influencing and correlating with the others.Simply applying these models sequentially can lead to error accumulation from each model, along with increased computational costs. To address these challenges, we introduce SuperPC, the first unified diffusion model capable of concurrently handling all four tasks. Our approach employs a three-level-conditioned diffusion framework, enhanced by a novel spatial-mix-fusion strategy, to leverage the correlations among these four defects for simultaneous, efficient processing.We show that SuperPC outperforms the state-of-the-art specialized models as well as their combination on all four individual tasks.

Multimodal 3D segmentation involves a significant number of 3D convolution operations, which requires substantial computational resources and high-performance computing devices in MRI multimodal brain tumor segmentation. The key challenge in multimodal 3D segmentation is how to minimize network computational load while maintaining high accuracy. To address the issue, a novel lightweight parameter aggregation network (SuperLightNet) is proposed to realize the efficient encoder and decoder for the high accurate and low computation. A random multiview drop encoder is designed to learn the spatial structure of multimodal images through a random multi-view approach for solving the high computational time complexity that has arisen in recent years with methods relying on transformers and Mamba. A learnable residual skip decoder is designed to incorporate learnable residual and group skip weights for addressing the reduced computational efficiency caused by the use of overly heavy convolution and deconvolution decoders. Experimental results demonstrate that the proposed method achieves a leading reduction in parameter count by 95.59%, the 96.78% improvement in computational efficiency, the 96.86% enhancement in memory access performance, and the average performance gain of 0.21% on the BraTS2019 and BraTS2021 datasets in comparison with the state-of-the-art methods. Code is available at https://github.com/WTU1020-Medical-Segmentation/SuperLightNet.

Learning robust models under distribution shifts between training and test datasets is a fundamental challenge in machine learning. While learning invariant features across environments is a popular approach, it often assumes that these features are fully observed in both training and test sets--a condition frequently violated in practice. When models rely on invariant features absent in the test set, their robustness in new environments can deteriorate. To tackle this problem, we introduce a novel learning principle called the Sufficient Invariant Learning (SIL) framework, which focuses on learning a sufficient subset of invariant features rather than relying on a single feature. After demonstrating the limitation of existing invariant learning methods, we propose a new algorithm, Adaptive Sharpness-aware Group Distributionally Robust Optimization (ASGDRO), to learn diverse invariant features by seeking common flat minima across the environments. We theoretically demonstrate that finding a common flat minima enables robust predictions based on diverse invariant features. Empirical evaluations on multiple datasets, including our new benchmark, confirm ASGDRO's robustness against distribution shifts, highlighting the limitations of existing methods.

Heterogeneous Federated Learning (HFL) has received widespread attention due to its adaptability to different models and data. The HFL approach utilizing auxiliary models for knowledge transfer enhances flexibility. However, existing frameworks face the challenges of aggregation bias and local overfitting. To address these issues, we propose FedSCE. It reduces the degree of freedom of update and improves generalization performance by limiting the specific layer of local model update to the local subspace. The subspace is dynamically updated to ensure coverage of the latest model update trajectory. Additionally, FedSCE evaluates client contributions based on the update distance of the auxiliary model in feature space and parameter space, achieving adaptive weighted aggregation. We validate our approach in both feature-skewed and label-skewed scenarios, demonstrating that on Office10, our method exceeds the best baseline by 3.87%. The code will be available at https://github.com/AVC2-UESTC/FedSCE.git.

N-shot neural architecture search (NAS) exploits a supernet containing all candidate subnets for a given search space. The subnets are typically trained with a static training strategy (e.g., using the same learning rate (LR) scheduler and optimizer for all subnets). This, however, does not consider that individual subnets have distinct characteristics, leading to two problems: (1) The supernet training is biased towards the low-complexity subnets (unfairness); (2) the momentum update in the supernet is noisy (noisy momentum). We present a dynamic supernet training technique to address these problems by adjusting the training strategy adaptive to the subnets. Specifically, we introduce a complexity-aware LR scheduler (CaLR) that controls the decay ratio of LR adaptive to the complexities of subnets, which alleviates the unfairness problem. We also present a momentum separation technique (MS). It groups the subnets with similar structural characteristics and uses a separate momentum for each group, avoiding the noisy momentum problem. Our approach can be applicable to various N-shot NAS methods with marginal cost, while improving the search performance drastically. We validate the effectiveness of our approach on various search spaces (e.g., NAS-Bench-201, Mobilenet spaces) and datasets (e.g., CIFAR-10/100, ImageNet).

Text-driven style transfer aims to merge the style of a reference image with content described by a text prompt. Recent advancements in text-to-image models have improved the nuance of style transformations, yet significant challenges remain, particularly with overfitting to reference styles, limiting stylistic control, and misaligning with textual content.In this paper, we propose three complementary strategies to address these issues. First, we introduce a cross-modal Adaptive Instance Normalization (AdaIN) mechanism for better integration of style and text features, enhancing alignment. Second, we develop a Style-based Classifier-Free Guidance (SCFG) approach that enables selective control over stylistic elements, reducing irrelevant influences. Finally, we incorporate a teacher model during early generation stages to stabilize spatial layouts and mitigate artifacts. Our extensive evaluations demonstrate significant improvements in style transfer quality and alignment with textual prompts. Furthermore, our approach can be integrated into existing style transfer frameworks without fine-tuning.

Style control has been popular in video generation models. Existing methods often generate videos far from the given style, cause content leakage, and struggle to transfer one video to the desired style. Our first observation is that the style extraction stage matters, whereas existing methods emphasize global style but ignore local textures. In order to bring texture features while preventing content leakage, we filter content-related patches while retaining style ones based on prompt-patch similarity; for global style extraction, we generate a paired style dataset through model illusion to facilitate contrastive learning, which greatly enhances the absolute style consistency. Moreover, to fill in the image-to-video gap, we train a lightweight motion adapter on still videos, which implicitly enhances stylization extent, and enables our image-trained model to be seamlessly applied to videos. Benefited from these efforts, our approach, StyleMaster, not only achieves significant improvement in both style resemblance and temporal coherence, but also can easily generalize to video style transfer with a gray tile ControlNet. Extensive experiments and visualizations demonstrate that StyleMaster significantly outperforms competitors, effectively generating high-quality stylized videos that align with textual content and closely resemble the style of reference images.

Under limited data setting, GANs often struggle to navigate and effectively exploit the input latent space. Consequently, images generated from adjacent variables in a sparse input latent space may exhibit significant discrepancies in realism, leading to suboptimal consistency regularization (CR) outcomes. To address this, we propose SQ-GAN, a novel approach that enhances CR by introducing a style space quantization scheme. This method transforms the sparse, continuous input latent space into a compact, structured discrete proxy space, allowing each element to correspond to a specific data point, thereby improving CR performance. Instead of direct quantization, we first map the input latent variables into a less entangled "style" space and apply quantization using a learnable codebook. This enables each quantized code to control distinct factors of variation. Additionally, we optimize the optimal transport distance to align the codebook codes with features extracted from the training data by a foundation model, embedding external knowledge into the codebook and establishing a semantically rich vocabulary that properly describes the training dataset. Extensive experiments demonstrate significant improvements in both discriminator robustness and generation quality with our method.

As scaling laws in generative AI push performance, they simultaneously concentrate the development of these models among actors with large computational resources. With a focus on text-to-image (T2I) generative models, we aim to unlock this bottleneck by demonstrating very low-cost training of large-scale T2I diffusion transformer models. As the computational cost of transformers increases with the number of patches in each image, we propose randomly masking up to 75% of the image patches during training. We propose a deferred masking strategy that preprocesses all patches using a patch-mixer before masking, thus significantly reducing the performance degradation with masking, making it superior to model downscaling in reducing computational cost. We also incorporate the latest improvements in transformer architecture, such as the use of mixture-of-experts layers, to improve performance and further identify the critical benefit of using synthetic images in micro-budget training. Finally, using only 37M publicly available real and synthetic images, we train a 1.16 billion parameter sparse transformer with only 1,890 USD economical cost and achieve a 12.7 FID in zero-shot generation on the COCO dataset. Notably, our model achieves competitive performance across both automated and human-centric evaluations, as well as high-quality generations, while incurring 118xlower costs than Stable Diffusion models and 14xlower costs than the current state-of-the-art approach, which costs \28,400. We also further investigate the influence of synthetic images on performance and demonstrate that micro-budget training on only synthetic images is sufficient for achieving high-quality data generation. Our end-to-end training pipeline and model checkpoints are available at https://github.com/SonyResearch/micro_diffusion.