CVPR 2025 Papers

On top of Segment Anything Model (SAM), SAM 2 further extends its capability from image to video inputs through a memory bank mechanism and obtains a remarkable performance compared with previous methods, making it a foundation model for video segmentation task. In this paper, we aim at making SAM 2 much more efficient so that it even runs on mobile devices while maintaining a comparable performance. Despite several works optimizing SAM for better efficiency, we find they are not sufficient for SAM 2 because they all focus on compressing the image encoder, while our benchmark shows that the newly introduced memory attention blocks are also the latency bottleneck. Given this observation, we propose EdgeTAM, which leverages a novel 2D Spatial Perceiver to reduce the computational cost. In particular, the proposed 2D Spatial Perceiver encodes the densely stored frame-level memories with a lightweight Transformer that contains a fixed set of learnable queries. Given that video segmentation is a dense prediction task, we find preserving the spatial structure of the memories is essential so that the queries are split into global-level and patch-level groups. We also propose a distillation pipeline that further improves the performance without inference overhead. As a result, EdgeTAM achieves 87.7, 70.0, 72.3, and 71.7 J&F on DAVIS 2017, MOSE, SA-V val, and SA-V test, while running at 16 FPS on iPhone 15 Pro Max. The code and models are available at https://github.com/facebookresearch/EdgeTAM.

Character customization, or 'face crafting,' is a vital feature in role-playing games (RPGs), enhancing player engagement by enabling the creation of personalized avatars. Existing automated methods often struggle with generalizability across diverse game engines due to their reliance on the intermediate constraints of specific image domain and typically support only one type of input, either text or image. To overcome these challenges, we introduce EasyCraft, an innovative end-to-end feedforward framework that automates character crafting by uniquely supporting both text and image inputs. Our approach employs a translator capable of converting facial images of any style into crafting parameters. We first establish a unified feature distribution in the translator's image encoder through self-supervised learning on a large-scale dataset, enabling photos of any style to be embedded into a unified feature representation. Subsequently, we map this unified feature distribution to crafting parameters specific to a game engine, a process that can be easily adapted to most game engines and thus enhances EasyCraft's generalizability. By integrating text-to-image techniques with our translator, EasyCraft also facilitates precise, text-based character crafting. EasyCraft's ability to integrate diverse inputs significantly enhances the versatility and accuracy of avatar creation. Extensive experiments on two RPG games demonstrate the effectiveness of our method, achieving state-of-the-art results and facilitating adaptability across various avatar engines.

We propose EVOlutionary Selector (EVOS), an efficient training paradigm for accelerating Implicit Neural Representation (INR). Unlike conventional INR training that feeds all samples through the neural network in each iteration, our approach restricts training to strategically selected points, reducing computational overhead by eliminating redundant forward passes.Specifically, we treat each sample as an individual in an evolutionary process, where only those fittest ones survive and merit inclusion in training, adaptively evolving with the neural network dynamics. While this is conceptually similar to Evolutionary Algorithms, their distinct objectives (selection for acceleration vs. iterative solution optimization) require a fundamental redefinition of evolutionary mechanisms for our context.In response, we design sparse fitness evaluation, frequency-guided crossover, and augmented unbiased mutation to comprise EVOS. These components respectively guide sample selection with reduced computational cost, enhance performance through frequency-domain balance, and mitigate selection bias from cached evaluation. Extensive experiments demonstrate that our method achieves approximately 48%-66% reduction in training time while ensuring superior convergence without additional cost, establishing state-of-the-art acceleration among recent sampling-based strategies.

Multi-view clustering represents one of the most established paradigms within the field of unsupervised learning and has witnessed a surge in popularity in recent years. View-pair form contrastive learning allows for consistently representing multiple views by maximizing mutual information between each two views. This approach permits multi-view clustering to discern consistent latent representations across multiple views. However, two significant issues emerge when this approach is considered. i)it is challenging to ascertain which two views are most appropriate for contrastive learning when there are more than three views, particularly without prior knowledge. ii) when all views are included in contrastive learning, multi-view clustering performance is compromised by poor quality views. To tackle these issues, we present a novel Efficient Dual Selection Mechanism for deep Multi-View Clustering framework, termed EASEMVC. Specifically, EASEMVC first constructs a view graph based on the OT distance between the bipartite graph of each view. It then designs a view selection module to realize an efficient view-level selection process through the view topology relations in the view graph structure. Additionally, a cross-view sample graph structure is constructed at the sample level, with the sample topological relations in the cross-view sample graph structure being employed to generate reliable sample learning weights. Based on the view pairs and sample weights selected by the aforementioned method, we employ contrastive learning within the theoretical framework of information theory to obtain consistent representations between views. Extensive experiments on six benchmark datasets have demonstrated that the proposed EASEMVC outperforms the state-of-the-art methods.

While deep neural networks have revolutionized image denoising capabilities, their deployment on edge devices remains challenging due to substantial computational and memory requirements. To this end, we present DnLUT, an ultra-efficient lookup table-based framework that achieves high-quality color image denoising with minimal resource consumption. Our key innovation lies in two complementary components: a Pairwise Channel Mixer (PCM) that effectively captures inter-channel correlations and spatial dependencies in parallel, and a novel L-shaped convolution design that maximizes receptive field coverage while minimizing storage overhead. By converting these components into optimized lookup tables post-training, DnLUT achieves remarkable efficiency - requiring only 500KB storage and 0.1% energy consumption compared to its CNN contestant DnCNN, while delivering 20x faster inference. Extensive experiments demonstrate that DnLUT outperforms all existing LUT-based methods by over 1dB in PSNR, establishing a new state-of-the-art in resource-efficient color image denoising.

Many real-world applications of flow-based generative models desire a diverse set of samples that cover multiple modes of the target distribution. However, the predominant approach for obtaining diverse sets is not sample-efficient, as it involves independently obtaining many samples from the source distribution and mapping them through the flow until the desired mode coverage is achieved. As an alternative to repeated sampling, we introduce DiverseFlow: a training-free approach to improve the diversity of flow models. Our key idea is to employ a determinantal point process to induce a coupling between the samples that drives diversity under a fixed sampling budget. In essence, DiverseFlow allows exploration of more variations in a learned flow model with fewer samples. We demonstrate the efficacy of our method for tasks where sample-efficient diversity is desirable, such as text-guided image generation with polysemous words, inverse problems like large-hole inpainting, and class-conditional image synthesis.

Upcycled Mixture-of-Experts (MoE) models have shown great potential in various tasks by converting the original Feed-Forward Network (FFN) layers in pre-trained dense models into MoE layers. However, these models still suffer from significant parameter inefficiency due to the introduction of multiple experts. In this work, we propose a novel DeRS (Decompose, Replace, and Synthesis) paradigm to overcome this shortcoming, which is motivated by our observations about the unique redundancy mechanisms of upcycled MoE experts. Specifically, DeRS decomposes the experts into one expert-shared base weight and multiple expert-specific delta weights, and subsequently represents these delta weights in lightweight forms. Our proposed DeRS paradigm can be applied to enhance parameter efficiency in two different scenarios, including: 1) DeRS Compression for inference stage, using sparsification or quantization to compress vanilla upcycled MoE models; and 2) DeRS Upcycling for training stage, employing lightweight sparse or low-rank matrixes to efficiently upcycle dense models into MoE models. Extensive experiments across three different tasks show that the proposed methods can achieve extreme parameter efficiency while maintaining the performance for both training and compression of upcycled MoE models.

Long Video Temporal Grounding (LVTG) aims at identifying specific moments within lengthy videos based on user-provided text queries for effective content retrieval. The approach taken by existing methods of dividing video into clips and processing each clip via a full-scale expert encoder is challenging to scale due to prohibitive computational costs of processing a large number of clips in long videos. To address this issue, we introduce DeCafNet, an approach employing "delegate-and-conquer" strategy to achieve computation efficiency without sacrificing grounding performance. DeCafNet introduces a sidekick encoder that performs dense feature extraction over all video clips in a resource-efficient manner, while generating a saliency map to identify the most relevant clips for full processing by the expert encoder. To effectively leverage features from sidekick and expert encoders that exist at different temporal resolutions, we introduce DeCaf-Grounder, which unifies and refines them via query-aware temporal aggregation and multi-scale temporal refinement for accurate grounding. Experiments on two LTVG benchmark datasets demonstrate that DeCafNet reduces computation by up to 47% while still outperforming existing methods, establishing a new state-of-the-art for LTVG in terms of both efficiency and performance.

Dataset distillation has emerged as a powerful approach for reducing data requirements in deep learning. Among various methods, distribution matching-based approaches stand out for their balance of computational efficiency and strong performance. However, existing distance metrics used in distribution matching often fail to accurately capture distributional differences, leading to unreliable measures of discrepancy. In this paper, we reformulate dataset distillation as a minmax optimization problem and introduce Neural Characteristic Function Discrepancy (NCFD), a comprehensive and theoretically grounded metric for measuring distributional differences. NCFD leverages the Characteristic Function (CF) to encapsulate full distributional information, employing a neural network to optimize the sampling strategy for the CF's frequency arguments, thereby maximizing the discrepancy to enhance distance estimation. Simultaneously, we minimize the difference between real and synthetic data under this optimized NCFD measure. Our approach, termed Neural Characteristic Function Matching (NCFM), inherently aligns the phase and amplitude of neural features in the complex plane for both real and synthetic data, achieving a balance between realism and diversity in synthetic samples. Experiments demonstrate that our method achieves significant performance gains over state-of-the-art methods on both low- and high-resolution datasets. Notably, we achieve a 22.9% accuracy boost on ImageSquawk. Our method also reduces GPU memory usage by over 300x and achieves 20x faster processing speeds compared to state-of-the-art methods. To the best of our knowledge, this is the first work to achieve lossless compression of CIFAR-100 on a single NVIDIA 2080 Ti GPU using only 2.3 GB of memory. The code for this work is publicly available at: https://github.com/gszfwsb/NCFM.

Thermal imaging is often compromised by dynamic, complex degradations caused by hardware limitations and unpredictable environmental factors. The scarcity of high-quality infrared data, coupled with the challenges of dynamic, intricate degradations, makes it difficult to recover details using existing methods. In this paper, we introduce thermal degradation simulation integrated into the training process via a mini-max optimization, by modeling these degraded factors as adversarial attacks on thermal images. The simulation is dynamic to maximize objective functions, thus capturing a broad spectrum of degraded data distributions. This approach enables training with limited data, thereby improving model performance. Additionally, we introduce a dual-interaction network that combines the benefits of spiking neural networks with scale transformation to capture degraded features with sharp spike signal intensities. This architecture ensures compact model parameters while preserving efficient feature representation. Extensive experiments demonstrate that our method not only achieves superior visual quality under diverse single and composited degradation, but also delivers a significant reduction in processing when trained on only fifty clear images, outperforming existing techniques in efficiency and accuracy. The source code will be available at https://github.com/LiuZhu-CV/DEAL.

In the rapidly advancing field of image generation, *Visual Auto-Regressive* (VAR) modeling has garnered considerable attention for its innovative next-scale prediction approach. This paradigm offers substantial improvements in efficiency, scalability, and zero-shot generalization. Yet, the inherently coarse-to-fine nature of VAR introduces a prolonged token sequence, leading to prohibitive memory consumption and computational redundancies. To overcome these bottlenecks, we propose *Collaborative Decoding* (CoDe), a novel decoding strategy tailored to the VAR framework. CoDe capitalizes on two critical observations: the substantially reduced parameter demands at larger scales and the exclusive generation patterns across different scales. Based on these insights, we partition the multi-scale inference process into a seamless collaboration between a large model and a small model. The large model serves as the 'drafter', specializing in generating low-frequency content at smaller scales, while the smaller model serves as the 'refiner', solely focusing on predicting high-frequency details at larger scales. This collaboration yields remarkable efficiency with minimal impact on quality: CoDe achieves a 1.7x speedup, slashes memory usage by around 50%, and preserves image quality with only a negligible FID increase from 1.95 to 1.98. When drafting steps are further decreased, CoDe can achieve an impressive 2.9x acceleration, reaching over 41 images/s at 256x256 resolution on a single NVIDIA 4090 GPU, while preserving a commendable FID of 2.27.

We address the task of video chaptering, i.e., partitioning a long video timeline into semantic units and generating corresponding chapter titles. While relatively underexplored, automatic chaptering has the potential to enable efficient navigation and content retrieval in long-form videos. In this paper, we achieve strong chaptering performance on hour-long videos by efficiently addressing the problem in the text domain with our "Chapter-Llama" framework. Specifically, we leverage a pre-trained large language model (LLM) with large context window, and feed as input (i) speech transcripts and (ii) captions describing video frames, along with their respective timestamps. Given the inefficiency of exhaustively captioning all frames, we propose a lightweight speech-guided frame selection strategy based on speech transcript content, and experimentally demonstrate remarkable advantages. We train the LLM to output timestamps for the chapter boundaries, as well as free-form chapter titles. This simple yet powerful approach scales to processing one-hour long videos in a single forward pass. Our results demonstrate substantial improvements (e.g., 45.3 vs 26.7 F1 score) over the state of the art on the recent VidChapters-7M benchmark. To promote further research, we release our code and models at our project page.

Training large-scale neural networks in vision, and multimodal domains demands substantial memory resources, primarily due to the storage of optimizer states. While LoRA, a popular parameter-efficient method, reduces memory usage, it often suffers from suboptimal performance due to the constraints of low-rank updates. Low-rank gradient projection methods (e.g., GaLore, Flora) reduce optimizer memory by projecting gradients and moment estimates into low-rank spaces via singular value decomposition or random projection. However, they fail to account for inter-projection correlation, causing performance degradation, and their projection strategies often incur high computational costs. In this paper, we present COAP (Correlation-Aware Gradient Projection), a memory-efficient method that minimizes computational overhead while maintaining training performance. Evaluated across various vision, language, and multimodal tasks, COAP outperforms existing methods in both training speed and model performance. For LLaMA-1B, it reduces optimizer memory by 61% with only 2% additional time cost, achieving the same PPL as AdamW. With 8-bit quantization, COAP cuts optimizer memory by 81% and achieves 4x speedup over GaLore for LLaVA-v1.5-7B fine-tuning, while delivering higher accuracy.

Scene Text Retrieval (STR) seeks to identify all images containing a given query string. Existing methods typically rely on an explicit Optical Character Recognition (OCR) process of text spotting or localization, which is susceptible to complex pipelines and accumulated errors. To settle this, we resort to the Contrastive Language-Image Pre-training (CLIP) models, which have demonstrated the capacity to perceive and understand scene text, making it possible to achieve strictly OCR-free STR. From the perspective of parameter-efficient transfer learning, a lightweight visual position adapter is proposed to provide a positional information complement for the visual encoder. Besides, we introduce a visual context dropout technique to improve the alignment of local visual features. A novel, parameter-free cross-attention mechanism transfers the contrastive relationship between images and text to that between visual tokens and text, producing a rich cross-modal representation, which can be utilized for efficient reranking with a linear classifier. The resulting model, CAYN, which proves that CLIP is Almost all You Need for STR with no more than 0.50M additional parameters required, achieves new state-of-the-art performance on the STR task, with 92.46%/89.49%/85.98% mAP on the SVT/IIIT-STR/TTR datasets. Our findings demonstrate that CLIP can serve as a reliable and efficient solution for OCR-free STR.

Transformer-based methods have demonstrated impressive performance in low-level visual tasks such as Image Super-Resolution (SR). However, its computational complexity grows quadratically with the spatial resolution. A series of works attempt to alleviate this problem by dividing Low-Resolution images into local windows, axial stripes, or dilated windows. SR typically leverages the redundancy of images for reconstruction, and this redundancy appears not only in local regions but also in long-range regions. However, these methods limit attention computation to content-agnostic local regions, limiting directly the ability of attention to capture long-range dependency. To address these issues, we propose a lightweight Content-Aware Token Aggregation Network (CATANet). Specifically, we propose an efficient Content-Aware Token Aggregation module for aggregate long-range content-similar tokens, which shares token centers across all image tokens and updates them only during the training phase. Then we utilize intra-group self-attention to enable long-range information interaction. Moreover, we design an inter-group cross-attention to further enhance global information interaction. The experimental results show that, compared with the state-of-the-art cluster-based method SPIN, our method achieves superior performance, with a maximum PSNR improvement of 0.33dB and nearly double the inference speed.

Brain-inspired spiking neural networks (SNNs) have the capability of energy-efficient processing of temporal information. However, leveraging the rich dynamic characteristics of SNNs and prior works in artificial neural networks (ANNs) to construct an effective object detection model for visual tasks remains an open question for further exploration. To develop a directly-trained , low energy consumption and high-performance multi-scale SNN model, we propose a novel interpretable object detection framework Multi-scale Spiking Detector (MSD). Initially, we propose a spiking convolutional neuron as a core component of the Optic Nerve Nucleus Block (ONNB), designed to significantly enhance the deep feature extraction capabilities of SNNs. ONNB enables direct training with improved energy efficiency, demonstrating superior performance compared to state-of-the-art ANN-to-SNN conversion and SNN techniques. In addition, we propose a Multi-scale Spiking Detection Framework to emulate the biological response and comprehension of stimuli from different objects. Wherein, spiking multi-scale fusion and the spiking detector are employed to integrate features across different depths and to detect response outcomes, respectively. Our method outperforms state-of-the-art ANN detectors, with only 7.8 M parameters and 6.43 mJ energy consumption. MSD obtains the mean average precision (mAP) of 62.0% and 66.3% on COCO and Gen1 datasets, respectively.

Federated learning (FL) enables collaborative model training with privacy preservation. Data heterogeneity across edge devices (clients) can cause models to converge to sharp minima, negatively impacting generalization and robustness. Recent approaches use client-side sharpness-aware minimization (SAM) to encourage flatter minima, but the discrepancy between local and global loss landscapes often undermines their effectiveness, as optimizing for local sharpness does not ensure global flatness. This work introduces FedGloSS (Federated Global Server-side Sharpness), a novel FL approach that prioritizes the optimization of global sharpness on the server, using SAM. To reduce communication overhead, FedGloSS cleverly approximates sharpness using the previous global gradient, eliminating the need for additional client communication. Our extensive evaluations demonstrate that FedGloSS consistently reaches flatter minima and better performance compared to state-of-the-art FL methods across various federated vision benchmarks. Code available at https://github.com/pietrocagnasso/fedgloss.

Structured pruning and quantization are fundamental techniques used to reduce the size of deep neural networks (DNNs) and typically are applied independently. Applying these techniques jointly via co-optimization has the potential to produce smaller, high-quality models. However, existing joint schemes are not widely used because of (1) engineering difficulties (complicated multi-stage processes), (2) black-box optimization (extensive hyperparameter tuning to control the overall compression), and (3) insufficient architecture generalization. To address these limitations, we present the framework GETA, which automatically and efficiently performs joint structured pruning and quantization-aware training on any DNN. GETA introduces three key innovations: (I) a quantization-aware dependency graph (QADG) that constructs a pruning search space for generic quantization-aware DNN, (II) a partially projected stochastic gradient method that guarantees layer-wise bit constraints are satisfied, and (III) a new joint learning strategy that incorporates interpretable relationships between pruning and quantization. We present numerical experiments on both convolutional neural networks and transformer architectures that show that our approach achieves competitive (often superior) performance compared to existing joint pruning and quantization methods. The source code is available at https://github.com/microsoft/GETA.

Self-Supervised Video Hashing (SSVH) compresses videos into hash codes for efficient indexing and retrieval using unlabeled training videos. Existing approaches rely on random frame sampling to learn video features and treat all frames equally. This results in suboptimal hash codes, as it ignores frame-specific information density and reconstruction difficulty. To address this limitation, we propose a new framework, termed AutoSSVH, that employs adversarial frame sampling with hash-based contrastive learning. Our adversarial sampling strategy automatically identifies and selects challenging frames with richer information for reconstruction, enhancing encoding capability. Additionally, we introduce a hash component voting strategy and a point-to-set (P2Set) hash-based contrastive objective, which help capture complex inter-video semantic relationships in the Hamming space and improve the discriminability of learned hash codes. Extensive experiments demonstrate that AutoSSVH achieves superior retrieval efficacy and efficiency compared to state-of-the-art approaches. Code is available at https://github.com/EliSpectre/CVPR25-AutoSSVH.