While foundation models have revolutionised computer vision, their effectiveness for sketch understanding remains limited by the unique challenges of abstract, sparse visual inputs. Through systematic analysis, we uncover two fundamental limitations: Stable Diffusion (SD) struggles to extract meaningful features from abstract sketches (unlike its success with photos), and exhibits a pronounced frequency-domain bias that suppresses essential low-frequency components needed for sketch understanding. Rather than costly retraining, we address these limitations by strategically combining SD with CLIP, whose strong semantic understanding naturally compensates for SD's spatial-frequency biases. By dynamically injecting CLIP features into SD's denoising process and adaptively aggregating features across semantic levels, our method achieves state-of-the-art performance in sketch retrieval (+3.35%), recognition (+1.06%), segmentation (+29.42%), and correspondence learning (+21.22%), demonstrating the first truly universal sketch feature representation in the era of foundation models.
论文检索
输入标题、作者或关键词，从 9,654 篇学术成果中精准定位
Boosting Point-Supervised Temporal Action Localization through Integrating Query Reformation and Optimal Transport
PDF ↗Point-supervised Temporal Action Localization poses significant challenges due to the difficulty of identifying complete actions with a single-point annotation per action. Existing methods typically employ Multiple Instance Learning, which struggles to capture global temporal context and requires heuristic post-processing. In research on fully-supervised tasks, DETR-based structures have effectively addressed these limitations. However, it is nontrivial to merely adapt DETR to this task, encountering two major bottlenecks. (1) How to integrate point label information into the model and (2) How to select optimal decoder proposals for training in the absence of complete action segment annotations. To address this issue, we introduce an end-to-end framework by integrating Query Reformation and Optimal Transport (QROT). Specifically, we encode point labels through a set of semantic consensus queries, enabling effective focus on action-relevant snippets. Furthermore, we integrate an optimal transport mechanism to generate high-quality pseudo labels. These pseudo-labels facilitate precise proposals selection based on Hungarian algorithm, significantly enhancing localization accuracy in point-supervised settings. Extensive experiments on the THUMOS14 and ActivityNet-v1.3 datasets demonstrate that our method outperforms existing MIL-based approaches, offering more stable and accurate temporal action localization in point-level supervision.
Subject-driven image personalization has seen notable advancements, especially with the advent of the ReferenceNet paradigm. ReferenceNet excels in integrating image reference features, making it highly applicable in creative and commercial settings. However, current implementations of ReferenceNet primarily operate as latent-level feature extractors, which limit their potential. This constraint hinders the provision of appropriate features to the denoising backbone across different timesteps, leading to suboptimal image consistency. In this paper, we revisit the extraction of reference features and propose TFCustom, a model framework designed to focus on reference image features at different temporal steps and frequency levels. Specifically, we firstly propose synchronized ReferenceNet to extract reference image features while simultaneously optimizing noise injection and denoising for the reference image. We also propose a time-aware frequency feature refinement module that leverages high- and low-frequency filters, combined with time embeddings, to adaptively select the degree of reference feature injection. Additionally, to enhance the similarity between reference objects and the generated image, we introduce a novel reward-based loss that encourages greater alignment between the reference and generated images. Experimental results demonstrate state-of-the-art performance in both multi-object and single-object reference generation, with significant improvements in texture and textual detail generation over existing methods.
Referring audio-visual segmentation (Ref-AVS) aims to segment objects within audio-visual scenes using multimodal cues embedded in text expressions. While the Segment Anything Model (SAM) has revolutionized visual segmentation, its applicability to Ref-AVS, where multimodal cues act as novel prompts, remains unexplored. SAM's limitation to single-frame segmentation also hinders its ability to capture essential temporal context needed for multi-frame audio-visual segmentation. To address this gap, we propose TSAM, a novel extension of SAM designed to leverage multimodal cues for precise segmentation in dynamic audio-visual scenes. TSAM enhances SAM's image encoder with a temporal modeling branch, enabling spatio-temporal learning and deep multimodal fusion across video frames, while retaining SAM's pre-trained knowledge. Additionally, TSAM replaces SAM's user-interactive prompting mechanism with sparse and dense data-driven prompts, enabling more effective integration of audio-visual inputs and reference text expressions. Extensive experiments on the Ref-AVS dataset demonstrate TSAM's superiority over state-of-the-art methods. The results illustrate its effectiveness in segmenting objects in dynamic audio-visual scenes using text-based multimodal cues and its strong generalization to unseen objects.
The rapid evolution of deep learning and large language models has led to an exponential growth in the demand for training data, prompting the development of Dataset Distillation methods to address the challenges of managing large datasets. Among these, Matching Training Trajectories (MTT) has been a prominent approach, which replicates the training trajectory of an expert network on real data with a synthetic dataset. However, our investigation found that this method suffers from three significant limitations: 1. Instability of expert trajectory generated by Stochastic Gradient Descent (SGD); 2. Low convergence speed of the distillation process; 3. High storage consumption of the expert trajectory. To address these issues, we offer a new perspective on understanding the essence of Dataset Distillation and MTT through a simple transformation of the objective function, and introduce a novel method called Matching Convexified Trajectory (MCT), which aims to provide better guidance for the student trajectory. MCT creates convex combinations of expert trajectories by selecting a few expert models, guiding student networks to converge quickly and stably. This trajectory is not only easier to store, but also enables continuous sampling strategies during the distillation process, ensuring thorough learning and fitting of the entire expert trajectory. The comprehensive experiment of three public datasets verified that MCT is superior to the traditional MTT method.
Recent advances in text-conditioned video diffusion have greatly improved video quality. However, these methods offer limited or sometimes no control to users on camera aspects, including dynamic camera motion, zoom, distorted lens and focus shifts. These motion and optical aspects are crucial for adding controllability and cinematic elements to generation frameworks, ultimately resulting in visual content that draws focus, enhances mood, and guides emotions according to filmmakers' controls. In this paper, we aim to close the gap between controllable video generation and camera optics. To achieve this, we propose AKiRa (Augmentation Kit on Rays), a novel augmentation framework that builds and trains a camera adapter with a complex camera model over an existing video generation backbone. It enables fine-tuned control over camera motion as well as complex optical parameters (focal length, distortion, aperture) to achieve cinematic effects such as zoom, fisheye effect, and bokeh. Extensive experiments demonstrate AKiRa's effectiveness in combining and composing camera optics while outperforming all state-of-the-art methods. This work sets a new landmark in controlled and optically enhanced video generation, paving the way for future camera diffusion methods.
We introduce a novel method for pre-training of large-scale vision encoders. Building on recent advancements in autoregressive pre-training of vision models, we extend this framework to a multimodal setting, i.e., images and text. In this paper, we present AIMV2, a family of generalist vision encoders characterized by a straightforward pre-training process, scalability, and remarkable performance across a range of downstream tasks. This is achieved by pairing the vision encoder with a multimodal decoder that autoregressively generates raw image patches and text tokens. Our encoders excel not only in multimodal evaluations but also in vision benchmarks such as localization, grounding, and classification. Notably, our AIMV2-3B encoder achieves 89.5% accuracy on ImageNet-1k with a frozen trunk. Fur- thermore, AIMV2 consistently outperforms state-of-the-art contrastive models (e.g., CLIP, SigLIP) in multimodal im- age understanding across diverse settings.
We study the problem of building a visual concept library for visual recognition. Building effective visual concept libraries is challenging, as manual definition is labor-intensive, while relying solely on LLMs for concept generation can result in concepts that lack discriminative power or fail to account for the complex interactions between them. Our approach, ESCHER, takes a library learning perspective to iteratively discover and improve visual concepts. ESCHER uses a vision-language model (VLM) as a critic to iteratively refine the concept library, including accounting for interactions between concepts and how they affect downstream classifiers. By leveraging the in-context learning abilities of LLMs and the history of performance using various concepts, ESCHER dynamically improves its concept generation strategy based on the VLM critic's feedback. Finally, ESCHER does not require any human annotations, and is thus an automated plug-and-play framework. We empirically demonstrate the ability of ESCHER to learn a concept library for zero-shot, few-shot, and fine-tuning visual classification tasks. This work represents, to our knowledge, the first application of concept library learning to real-world visual tasks.
Text-to-video (T2V) generative models have advanced significantly, yet their ability to compose different objects, attributes, actions, and motions into a video remains unexplored. Previous text-to-video benchmarks also neglect this important ability for evaluation. In this work, we conduct the first systematic study on compositional text-to-video generation. We propose T2V-CompBench, the first benchmark tailored for compositional text-to-video generation. T2V-CompBench encompasses diverse aspects of compositionality, including consistent attribute binding, dynamic attribute binding, spatial relationships, motion binding, action binding, object interactions, and generative numeracy. We further carefully design evaluation metrics of multimodal large language model (MLLM)-based, detection-based, and tracking-based metrics, which can better reflect the compositional text-to-video generation quality of seven proposed categories with 1400 text prompts. The effectiveness of the proposed metrics is verified by correlation with human evaluations. We also benchmark various text-to-video generative models and conduct in-depth analysis across different models and various compositional categories. We find that compositional text-to-video generation is highly challenging for current models, and we hope our attempt could shed light on future research in this direction.
MMTL-UniAD: A Unified Framework for Multimodal and Multi-Task Learning in Assistive Driving Perception
PDF ↗Advanced driver assistance systems require a comprehensive understanding of the driver's mental/physical state and traffic context but existing works often neglect the potential benefits of joint learning between these tasks. This paper proposes MMTL-UniAD, a unified multi-modal multi-task learning framework that simultaneously recognizes driver behavior (e.g., looking around, talking), driver emotion (e.g., anxiety, happiness), vehicle behavior (e.g., parking, turning), and traffic context (e.g., traffic jam, traffic smooth). A key challenge is avoiding negative transfer between tasks, which can impair learning performance. To address this, we introduce two key components into the framework: one is the multi-axis region attention network to extract global context-sensitive features, and the other is the dual-branch multimodal embedding to learn multimodal embeddings from both task-shared and task-specific features. The former uses a multi-attention mechanism to extract task-relevant features, mitigating negative transfer caused by task-unrelated features. The latter employs a dual-branch structure to adaptively adjust task-shared and task-specific parameters, enhancing cross-task knowledge transfer while reducing task conflicts. We assess MMTL-UniAD on the AIDE dataset, using a series of ablation studies, and show that it outperforms state-of-the-art methods across all four tasks. The code is available on https://github.com/Wenzhuo-Liu/MMTL-UniAD.
Interactive drag editing of images is a valuable task that has gained considerable attention for its precision and controllability. However, existing approaches have primarily focused on manipulating the shape or movement of objects in 2D plane. We propose to extend this drag-based editing task to 3D space. Firstly, we utilize the trajectory of two points to represent the rotational trajectory of the object. Gaussian maps of a circle and a square are centered at these two points, respectively. We use distinct shapes to ensure that symmetric views produce different object representations. Secondly, we introduce a lightweight mapping network to embed the object features into two Gaussian maps to obtain a continuous control condition that guides the model in learning the correspondence between the trajectory and the object. Finally, to overcome the limitations of current 3D object reconstruction datasets, which typically consist of object maps with transparent backgrounds, we affix random backgrounds to them. This modification helps improve the model's ability to ignore background interference when editing real images with complex backgrounds. Experiments demonstrate that our approach successfully achieves object rotation within the drag framework and demonstrates strong generalization to real-world images.
Understanding functionalities in 3D scenes involves interpreting natural language descriptions to locate functional interactive objects, such as handles and buttons, in a 3D environment. Functionality understanding is highly challenging, as it requires both world knowledge to interpret language and spatial perception to identify fine-grained objects. For example, given a task like "turn on the ceiling light", an embodied AI agent must infer that it needs to locate the light switch, even though the switch is not explicitly mentioned in the task description.To date, no dedicated methods have been developed for this problem. In this paper, we introduce Fun3DU, the first approach designed for functionality understanding in 3D scenes. Fun3DU uses a language model to parse the task description through Chain-of-Thought reasoning in order to identify the object of interest. The identified object is segmented across multiple views of the captured scene by using a VLM. The segmentation results from each view are lifted in 3D and aggregated into the point cloud using geometric information. Fun3DU is training-free, relying entirely on pre-trained models. We evaluate Fun3DU on SceneFun3D, the most recent and only dataset to benchmark this task, which comprises over 3000 task descriptions on 230 scenes. Our method significantly outperforms state-of-the-art open-vocabulary 3D segmentation approaches. Project page: https://tev-fbk.github.io/fun3du
Task Arithmetic has emerged as a simple yet effective method to merge models without additional training. However, by treating entire networks as flat parameter vectors, it overlooks key structural information and is susceptible to task interference. In this paper, we study task vectors at the layer level, focusing on task layer matrices and their singular value decomposition. In particular, we concentrate on the resulting singular vectors, which we refer to as Task Singular Vectors (TSV). Recognizing that layer task matrices are often low-rank, we propose TSV-Compress, a simple procedure that compresses them to 10% of their original size while retaining 99% of accuracy. We further leverage this low-rank space to define a new measure of task interference based on the interaction of singular vectors from different tasks. Building on these findings, we introduce TSV-Merge, a novel model merging approach that combines compression with interference reduction, significantly outperforming existing methods.
Diffusion models have demonstrated impressive performance in face restoration. Yet, their multi-step inference process remains computationally intensive, limiting their applicability in real-world scenarios. Moreover, existing methods often struggle to generate face images that are harmonious, realistic, and consistent with the subject's identity. In this work, we propose OSDFace, a novel one-step diffusion model for face restoration. Specifically, we propose a visual representation embedder (VRE) to better capture prior information and understand the input face. In VRE, low-quality faces are processed by a visual tokenizer and subsequently embedded with a vector-quantized dictionary to generate visual prompts. Additionally, we incorporate a facial identity loss derived from face recognition to further ensure identity consistency. We further employ a generative adversarial network (GAN) as a guidance model to encourage distribution alignment between the restored face and the ground truth. Experimental results demonstrate that OSDFace surpasses current state-of-the-art (SOTA) methods in both visual quality and quantitative metrics, generating high-fidelity, natural face images with high identity consistency. The code and model will be released at https://github.com/jkwang28/OSDFace.
Current structural pruning methods face two significant limitations: (i) they often limit pruning to finer-grained levels like channels, making aggressive parameter reduction challenging, and (ii) they focus heavily on parameter and FLOP reduction, with existing latency-aware methods frequently relying on simplistic, suboptimal linear models that fail to generalize well to transformers, where multiple interacting dimensions impact latency. In this paper, we address both limitations by introducing Multi-Dimensional Pruning(MDP), a novel paradigm that jointly optimizes across a variety of pruning granularities--including channels, query/key, heads, embeddings, and blocks. MDP employs an advanced latency modeling technique to accurately capture latency variations across all prunable dimensions, achieving an optimal balance between latency and accuracy. By reformulating pruning as a Mixed-Integer Nonlinear Program (MINLP), MDP efficiently identifies the optimal pruned structure across all prunable dimensions while respecting latency constraints. This versatile framework supports both CNNs and transformers. Extensive experiments demonstrate that MDP significantly outperforms previous methods, especially at high pruning ratios. On ImageNet, MDP achieves a 28% speed increase with a +1.4 Top-1 accuracy improvement over prior work like HALP for ResNet50 pruning. Against the latest transformer pruning method, Isomorphic, MDP delivers an additional 37% acceleration with a +0.7 Top-1 accuracy improvement.
Recently, learning signed distance functions (SDFs) from point clouds has become popular for reconstruction. To ensure accuracy, most methods require using high-resolution Marching Cubes for surface extraction. However, this results in redundant mesh elements, making the mesh inconvenient to use. To solve the problem, we propose an adaptive meshing method to extract resolution-adaptive meshes based on surface curvature, enabling the recovery of high-fidelity lightweight meshes. Specifically, we first use point-based representation to perceive implicit surfaces and calculate surface curvature. A vertex generator is designed to produce curvature-adaptive vertices with any specified number on the implicit surface, preserving the overall structure and high-curvature features. Then we develop a Delaunay meshing algorithm to generate meshes from vertices, ensuring geometric fidelity and correct topology. In addition, to obtain accurate SDFs for adaptive meshing and achieve better lightweight reconstruction, we design a hybrid representation combining feature grid and feature tri-plane for better detail capture. Experiments demonstrate that our method can generate high-quality lightweight meshes from point clouds. Compared with methods from various categories, our approach achieves superior results, especially in capturing more details with fewer elements.
DriveGPT4-V2: Harnessing Large Language Model Capabilities for Enhanced Closed-Loop Autonomous Driving
PDF ↗Multimodal large language models (MLLMs) possess the ability to comprehend visual images or videos, and show impressive reasoning ability thanks to the vast amounts of pretrained knowledge, making them highly suitable for autonomous driving applications. Unlike the previous work, DriveGPT4-V1, which focused on open-loop tasks, this study explores the capabilities of LLMs in enhancing closed-loop autonomous driving. DriveGPT4-V2 processes camera images and vehicle states as input to generate low-level control signals for end-to-end vehicle operation. A multi-view visual tokenizer (MV-VT) is employed enabling DriveGPT4-V2 to perceive the environment with an extensive range while maintaining critical details. The model architecture has been refined to improve decision prediction and inference speed. To further enhance the performance, an additional expert LLM is trained for online imitation learning. The expert LLM, sharing a similar structure with DriveGPT4-V2, can access privileged information about surrounding objects for more robust and reliable predictions. Experimental results show that DriveGPT4-V2 outperforms all baselines on the challenging CARLA Longest6 benchmark. The code and data of DriveGPT4-V2 will be publicly available.
Today's most advanced vision-language models (VLMs) remain proprietary. The strongest open-weight models rely heavily on synthetic data from proprietary VLMs to achieve good performance, effectively distilling these closed VLMs into open ones. As a result, the community has been missing foundational knowledge about how to build performant VLMs from scratch. We present Molmo, a new family of VLMs that are state-of-the-art in their class of openness. Our key contribution is a collection of new datasets called PixMo, including a dataset of highly detailed image captions for pre-training, a free-form image Q&A dataset for fine-tuning, and an innovative 2D pointing dataset, all collected without the use of external VLMs. The success of our approach relies on careful modeling choices, a well-tuned training pipeline, and, most critically, the quality of our newly collected datasets. Our best-in-class 72B model not only outperforms others in the class of open weight and data models, but also outperforms larger proprietary models including Claude 3.5 Sonnet, and Gemini 1.5 Pro and Flash, second only to GPT-4o based on both academic benchmarks and on a large human evaluation. Our model weights, new datasets, and source code are available at https://molmo.allenai.org/blog.
Pairwise pose estimation from images with little or no overlap is an open challenge in computer vision. Existing methods, even those trained on large-scale datasets, struggle in these scenarios due to the lack of identifiable correspondences or visual overlap. Inspired by the human ability to infer spatial relationships from diverse scenes, we propose a novel approach, InterPose, that leverages the rich priors encoded within pre-trained generative video models. We propose to use a video model to hallucinate intermediate frames between two input images, effectively creating a dense, visual transition, which significantly simplifies the problem of pose estimation. Since current video models can still produce implausible motion or inconsistent geometry, we introduce a self-consistency score that evaluates the consistency of pose predictions from sampled videos. We demonstrate that our approach generalizes among three state-of-the-art video models and show consistent improvements over the state-of-the-art DUSt3R baseline on four diverse datasets encompassing indoor, outdoor, and object-centric scenes. Our findings suggest a promising avenue for improving pose estimation models by leveraging large generative models trained on vast amounts of video data, which is more readily available than 3D data. See our project page for results: Inter-Pose.github.io.
Auxiliary-free human video matting methods, which rely solely on input frames, often struggle with complex or ambiguous backgrounds. To tackle this, we propose MatAnyone, a practical framework designed for target-assigned video matting. Specifically, building on a memory-based framework, we introduce a consistent memory propagation module via region-adaptive memory fusion, which adaptively combines memory from the previous frame. This ensures stable semantic consistency in core regions while maintaining fine details along object boundaries. For robust training, we present a larger, high-quality, and diverse dataset for video matting. Additionally, we incorporate a novel training strategy that efficiently leverages large-scale segmentation data, further improving matting stability. With this new network design, dataset, and training strategy, MatAnyone delivers robust, accurate video matting in diverse real-world scenarios, outperforming existing methods. The code and model will be publicly available.