ICML 2025 论文列表

Learning lighting adaptation is a crucial step in achieving good visual perception and supporting downstream vision tasks. Current research often addresses individual light-related challenges, such as high dynamic range imaging and exposure correction, in isolation. However, we identify shared fundamental properties across these tasks: i) different color channels have different light properties, and ii) the channel differences reflected in the spatial and frequency domains are different. Leveraging these insights, we introduce the channel-aware Learning Adaptive Lighting Network (LALNet), a multi-task framework designed to handle multiple light-related tasks efficiently. Specifically, LALNet incorporates color-separated features that highlight the unique light properties of each color channel, integrated with traditional color-mixed features by Light Guided Attention (LGA). The LGA utilizes color-separated features to guide color-mixed features focusing on channel differences and ensuring visual consistency across all channels. Additionally, LALNet employs dual domain channel modulation for generating color-separated features and a mixed channel modulation and light state space module for producing color-mixed features. Extensive experiments on four representative light-related tasks demonstrate that LALNet significantly outperforms state-of-the-art methods on benchmark tests and requires fewer computational resources. We provide an anonymous online demo at [LALNet](https://xxxxxx2025.github.io/LALNet/).

Recent advances in vision-language models have led to impressive progress in caption generation for images and short video clips. However, these models remain constrained by their limited temporal receptive fields, making it difficult to produce coherent and comprehensive captions for long videos. While several methods have been proposed to aggregate information across video segments, they often rely on supervised fine-tuning or incur significant computational overhead. To address these challenges, we introduce a novel framework for long video captioning based on graph consolidation. Our approach first generates segment-level captions, corresponding to individual frames or short video intervals, using off-the-shelf visual captioning models. These captions are then parsed into individual scene graphs, which are subsequently consolidated into a unified graph representation that preserves both holistic context and fine-grained details throughout the video. A lightweight graph-to-text decoder then produces the final video-level caption. This framework effectively extends the temporal understanding capabilities of existing models without requiring any additional fine-tuning on long video datasets. Experimental results show that our method significantly outperforms existing LLM-based consolidation approaches, achieving strong zero-shot performance while substantially reducing computational costs.

Recent advances in conditional diffusion models have shown promise for generating realistic TalkingFace videos, yet challenges persist in achieving consistent head movement, synchronized facial expressions, and accurate lip synchronization over extended generations. To address these, we introduce the \textbf{M}otion-priors \textbf{C}onditional \textbf{D}iffusion \textbf{M}odel (\textbf{MCDM}), which utilizes both archived and current clip motion priors to enhance motion prediction and ensure temporal consistency. The model consists of three key elements: (1) an archived-clip motion-prior that incorporates historical frames and a reference frame to preserve identity and context; (2) a present-clip motion-prior diffusion model that captures multimodal causality for accurate predictions of head movements, lip sync, and expressions; and (3) a memory-efficient temporal attention mechanism that mitigates error accumulation by dynamically storing and updating motion features. We also introduce the {TalkingFace-Wild} dataset, a multilingual collection of over 200 hours of footage across 10 languages. Experimental results demonstrate the effectiveness of MCDM in maintaining identity and motion continuity for long-term TalkingFace generation.

Due to the challenges of processing temporal information, most trackers depend solely on visual discriminability and overlook the unique temporal coherence of video data. In this paper, we propose a lightweight and plug-and-play motion prompt tracking method. It can be easily integrated into existing vision-based trackers to build a joint tracking framework leveraging both motion and vision cues, thereby achieving robust tracking through efficient prompt learning. A motion encoder with three different positional encodings is proposed to encode the long-term motion trajectory into the visual embedding space, while a fusion decoder and an adaptive weight mechanism are designed to dynamically fuse visual and motion features. We integrate our motion module into three different trackers with five models in total. Experiments on seven challenging tracking benchmarks demonstrate that the proposed motion module significantly improves the robustness of vision-based trackers, with minimal training costs and negligible speed sacrifice. Code is available at https://github.com/zj5559/Motion-Prompt-Tracking.

Structured image understanding, such as interpreting tables and charts, requires strategically refocusing across various structures and texts within an image, forming a reasoning sequence to arrive at the final answer. However, current multimodal large language models (LLMs) lack this multihop selective attention capability. In this work, we introduce ReFocus, a simple yet effective framework that equips multimodal LLMs with the ability to generate ``visual thoughts'' by performing visual editing on the input image through code, shifting and refining their visual focuses. Specifically, ReFocus enables multimodal LLMs to generate Python codes to call tools and modify the input image, sequentially drawing boxes, highlighting sections, and masking out areas, thereby enhancing the visual reasoning process. We experiment upon a wide range of structured image understanding tasks involving tables and charts. ReFocus largely improves performance on all tasks over GPT-4o without visual editing, yielding an average gain of 11.0% on table tasks and 6.8% on chart tasks. We present an in-depth analysis of the effects of different visual edits, and reasons why ReFocus can improve the performance without introducing additional information. Further, we collect a 14k training set using ReFocus, and prove that such visual chain-of-thought with intermediate information offers a better supervision than standard VQA data, reaching a 8.0% average gain over the same model trained with QA pairs and 2.6% over CoT.

We present TUMTraf VideoQA, a novel dataset and benchmark designed for spatio-temporal video understanding in complex roadside traffic scenarios. The dataset comprises 1,000 videos, featuring 85,000 multiple-choice QA pairs, 2,300 object captioning, and 5,700 object grounding annotations, encompassing diverse real-world conditions such as adverse weather and traffic anomalies. By incorporating tuple-based spatio-temporal object expressions, TUMTraf VideoQA unifies three essential tasks—multiple-choice video question answering, referred object captioning, and spatio-temporal object grounding—within a cohesive evaluation framework. We further introduce the TraffiX-Qwen baseline model, enhanced with visual token sampling strategies, providing valuable insights into the challenges of fine-grained spatio-temporal reasoning. Extensive experiments demonstrate the dataset’s complexity, highlight the limitations of existing models, and position TUMTraf VideoQA as a robust foundation for advancing research in intelligent transportation systems. The dataset and benchmark are publicly available to facilitate further exploration.

With the advance of diffusion models, today's video generation has achieved impressive quality. To extend the generation length and facilitate real-world applications, a majority of video diffusion models (VDMs) generate videos in an autoregressive manner, i.e., generating subsequent clips conditioned on the last frame(s) of the previous clip. However, existing autoregressive VDMs are highly inefficient and redundant: The model must re-compute all the conditional frames that are overlapped between adjacent clips. This issue is exacerbated when the conditional frames are extended autoregressively to provide the model with long-term context. In such cases, the computational demands increase significantly (i.e., with a quadratic complexity w.r.t. the autoregression step). In this paper, we propose **Ca2-VDM**, an efficient autoregressive VDM with **Ca**usal generation and **Ca**che sharing. For **causal generation**, it introduces unidirectional feature computation, which ensures that the cache of conditional frames can be precomputed in previous autoregression steps and reused in every subsequent step, eliminating redundant computations. For **cache sharing**, it shares the cache across all denoising steps to avoid the huge cache storage cost. Extensive experiments demonstrated that our Ca2-VDM achieves state-of-the-art quantitative and qualitative video generation results and significantly improves the generation speed. Code is available: https://github.com/Dawn-LX/CausalCache-VDM

On-device learning has emerged as a promising direction for AI development, particularly because of its potential to reduce latency issues and mitigate privacy risks associated with device-server communication, while improving energy efficiency. Despite these advantages, significant memory and computational constraints still represent major challenges for its deployment. Drawing on previous studies on low-rank decomposition methods that address activation memory bottlenecks in backpropagation, we propose a novel shortcut approach as an alternative. Our analysis and experiments demonstrate that our method can reduce activation memory usage, even up to $120.09\times$ compared to vanilla training, while also reducing overall training FLOPs up to $1.86\times$ when evaluated on traditional benchmarks.

Satellite imagery is essential for Earth observation, enabling applications like crop yield prediction, environmental monitoring, and climate change assessment. However, integrating satellite imagery with climate data remains a challenge, limiting its utility for forecasting and scenario analysis. We introduce a novel dataset of 2.9 million Sentinel-2 images spanning 15 land cover types with corresponding climate records, forming the foundation for two satellite image generation approaches using fine-tuned Stable Diffusion 3 models. The first is a text-to-image generation model that uses textual prompts with climate and land cover details to produce realistic synthetic imagery for specific regions. The second leverages ControlNet for multi-conditional image generation, preserving spatial structures while mapping climate data or generating time-series to simulate landscape evolution. By combining synthetic image generation with climate and land cover data, our work advances generative modeling in remote sensing, offering realistic inputs for environmental forecasting and new possibilities for climate adaptation and geospatial analysis.

We introduce a novel resampling criterion using lift scores, for improving compositional generation in diffusion models. By leveraging the lift scores, we evaluate whether generated samples align with each single condition and then compose the results to determine whether the composed prompt is satisfied. Our key insight is that lift scores can be efficiently approximated using only the original diffusion model, requiring no additional training or external modules. We develop an optimized variant that achieves relatively lower computational overhead during inference while maintaining effectiveness. Through extensive experiments, we demonstrate that lift scores significantly improved the condition alignment for compositional generation across 2D synthetic data, CLEVR position tasks, and text-to-image synthesis. Our code is available at github.com/rainorangelemon/complift.

N:M sparsity stands as a progressively important tool for DNN compression, achieving practical speedups by stipulating at most N non-zero components within M sequential weights. Unfortunately, most existing works identify the N:M sparse mask through dense backward propagation to update all weights, which incurs exorbitant training costs. In this paper, we introduce BAME, a method that maintains consistent sparsity throughout the N:M sparse training process. BAME perpetually keeps both sparse forward and backward propagation, while iteratively performing weight pruning-and-regrowing within designated weight blocks to tailor the N:M mask. These blocks are selected through a joint assessment based on accumulated mask oscillation frequency and expected loss reduction of mask adaptation, thereby ensuring stable and efficient identification of the optimal N:M mask. Our empirical results substantiate the effectiveness of BAME, illustrating it performs comparably to or better than previous works that fully maintaining dense backward propagation during training. For instance, BAME attains a 72.0% top-1 accuracy while training a 1:16 sparse ResNet-50 on ImageNet, eclipsing SR-STE by 0.5%, despite achieving 2.37 training FLOPs reduction. Code is released at \url{https://github.com/BAME-xmu/BAME}

Event-based object detection has attracted increasing attention for its high temporal resolution, wide dynamic range, and asynchronous address-event representation. Leveraging these advantages, spiking neural networks (SNNs) have emerged as a promising approach, offering low energy consumption and rich spatiotemporal dynamics. To further enhance the performance of event-based object detection, this study proposes a novel hybrid spike vision Transformer (HsVT) model. The HsVT model integrates a spatial feature extraction module to capture local and global features, and a temporal feature extraction module to model time dependencies and long-term patterns in event sequences. This combination enables HsVT to capture spatiotemporal features, improving its capability in handling complex event-based object detection tasks. To support research in this area, we developed the Fall Detection dataset as a benchmark for event-based object detection tasks. The Fall DVS detection dataset protects facial privacy and reduces memory usage thanks to its event-based representation. Experimental results demonstrate that HsVT outperforms existing SNN methods and achieves competitive performance compared to ANN-based models, with fewer parameters and lower energy consumption.

This paper presents a novel approach to hyperspectral image (HSI) reconstruction from RGB images, addressing fundamental limitations in existing learning-based methods from a physical perspective. We discuss and aim to address the ``colorimetric dilemma": failure to consistently reproduce ground-truth RGB from predicted HSI, thereby compromising physical integrity and reliability in practical applications. To tackle this issue, we propose PhySpec, a physically consistent framework for robust HSI reconstruction. Our approach fundamentally exploits the intrinsic physical relationship between HSIs and corresponding RGBs by employing orthogonal subspace decomposition, which enables explicit estimation of camera spectral sensitivity (CSS). This ensures that our reconstructed spectra align with well-established physical principles, enhancing their reliability and fidelity. Moreover, to efficiently use internal information from test samples, we propose a self-supervised meta-auxiliary learning (MAXL) strategy that rapidly adapts the trained parameters to unseen samples using only a few gradient descent steps at test time, while simultaneously constraining the generated HSIs to accurately recover ground-truth RGB values. Thus, MAXL reinforces the physical integrity of the reconstruction process. Extensive qualitative and quantitative evaluations validate the efficacy of our proposed framework, showing superior performance compared to SOTA methods.

Text-to-3D generation automates 3D content creation from textual descriptions, which offers transformative potential across various fields. However, existing methods often struggle to align generated content with human preferences, limiting their applicability and flexibility. To address these limitations, in this paper, we propose DreamDPO, an optimization-based framework that integrates human preferences into the 3D generation process, through direct preference optimization. Practically, DreamDPO first constructs pairwise examples, then validates their alignment with human preferences using reward or large multimodal models, and lastly optimizes the 3D representation with a preference-driven loss function. By leveraging relative preferences, DreamDPO reduces reliance on precise quality evaluations while enabling fine-grained controllability through preference-guided optimization. Experiments demonstrate that DreamDPO achieves state-of-the-art results, and provides higher-quality and more controllable 3D content compared to existing methods. The code and models will be open-sourced.

Accuracy and speed are critical in image editing tasks. Pan et al. introduced a drag-based framework using Generative Adversarial Networks, and subsequent studies have leveraged large-scale diffusion models. However, these methods often require over a minute per edit and exhibit low success rates. We present LightningDrag, which achieves high-quality drag-based editing in about one second on general images. By redefining drag-based editing as a conditional generation task, we eliminate the need for time-consuming latent optimization or gradient-based guidance. Our model is trained on large-scale paired video frames, capturing diverse motion (object translations, pose shifts, zooming, etc.) to significantly improve accuracy and consistency. Despite being trained only on videos, our model generalizes to local deformations beyond the training data (e.g., lengthening hair, twisting rainbows). Extensive evaluations confirm the superiority of our approach, and we will release both code and model.

The Segment Anything Model (SAM), a vision foundation model, exhibits impressive zero-shot capabilities in general tasks but struggles in specialized domains. Parameter-efficient fine-tuning (PEFT) is a promising approach to unleash the potential of SAM in novel scenarios. However, existing PEFT methods for SAM neglect the domain-invariant relations encoded in the pre-trained model. To bridge this gap, we propose InfoSAM, an information-theoretic approach that enhances SAM fine-tuning by distilling and preserving its pre-trained segmentation knowledge. Specifically, we formulate the knowledge transfer process as two novel mutual information-based objectives: (i) to compress the domain-invariant relation extracted from pre-trained SAM, excluding pseudo-invariant information as possible, and (ii) to maximize mutual information between the relational knowledge learned by the teacher (pre-trained SAM) and the student (fine-tuned model). The proposed InfoSAM establishes a robust distillation framework for PEFT of SAM. Extensive experiments across diverse benchmarks validate InfoSAM's effectiveness in improving SAM family's performance on real-world tasks, demonstrating its adaptability and superiority in handling specialized scenarios. The code and models are available at https://muyaoyuan.github.io/InfoSAM_Page.

We consider the problem of novel view synthesis from unposed images in a single feed-forward. Our framework capitalizes on fast speed, scalability, and high-quality 3D reconstruction and view synthesis capabilities of 3DGS, where we further extend it to offer a practical solution that relaxes common assumptions such as dense image views, accurate camera poses, and substantial image overlaps. We achieve this through identifying and addressing unique challenges arising from the use of pixel-aligned 3DGS: misaligned 3D Gaussians across different views induce noisy or sparse gradients that destabilize training and hinder convergence, especially when above assumptions are not met. To mitigate this, we employ pre-trained monocular depth estimation and visual correspondence models to achieve coarse alignments of 3D Gaussians. We then introduce lightweight, learnable modules to refine depth and pose estimates from the coarse alignments, improving the quality of 3D reconstruction and novel view synthesis. Furthermore, the refined estimates are leveraged to estimate geometry confidence scores, which assess the reliability of 3D Gaussian centers and condition the prediction of Gaussian parameters accordingly. Extensive evaluations on large-scale real-world datasets demonstrate that PF3plat sets a new state-of-the-art across all benchmarks, supported by comprehensive ablation studies validating our design choices. We will make the code and weights publicly available.

Face Recognition (FR) tasks have made significant progress with the advent of Deep Neural Networks, particularly through margin-based triplet losses that embed facial images into high-dimensional feature spaces. During training, these contrastive losses focus exclusively on identity information as labels. However, we observe a multiscale geometric structure emerging in the embedding space, influenced by interpretable facial (e.g., hair color) and image attributes (e.g., contrast). We propose a geometric approach to describe the dependence or invariance of FR models to these attributes and introduce a physics-inspired alignment metric. We evaluate the proposed metric on controlled, simplified models and widely used FR models fine-tuned with synthetic data for targeted attribute augmentation. Our findings reveal that the models exhibit varying degrees of invariance across different attributes, providing insight into their strengths and weaknesses and enabling deeper interpretability. Code available here: https://github.com/mantonios107/attrs-fr-embs.

The trade-off between accuracy and interpretability has long been a challenge in machine learning (ML). This tension is particularly significant for emerging *interpretable-by-design* methods, which aim to redesign ML algorithms for trustworthy interpretability but often sacrifice accuracy in the process. In this paper, we address this gap by investigating the impact of deviations in concept representations—an essential component of interpretable models—on prediction performance and propose a novel framework to mitigate these effects. The framework builds on the principle of optimizing concept embeddings under constraints that preserve interpretability. Using a generative model as a test-bed, we rigorously prove that our algorithm achieves zero loss while progressively enhancing the interpretability of the resulting model. Additionally, we evaluate the practical performance of our proposed framework in generating explainable predictions for image classification tasks across various benchmarks. Compared to existing explainable methods, our approach not only improves prediction accuracy while preserving model interpretability across various large-scale benchmarks but also achieves this with significantly lower computational cost.