ACM MM 2024 论文列表

3D open-vocabulary scene understanding, crucial for advancing augmented reality and robotic applications, involves interpreting and locating specific regions within a 3D space as directed by natural language instructions. To this end, we introduce GOI, a framework that integrates semantic features from 2D vision-language foundation models into 3D Gaussian Splatting (3DGS) and identifies 3D Gaussians of Interest using an Optimizable Semantic-space Hyperplane. Our approach includes an efficient compression method that utilizes scene priors to condense noisy high-dimensional semantic features into compact low-dimensional vectors, which are subsequently embedded in 3DGS. During the open-vocabulary querying process, we adopt a distinct approach compared to existing methods, which depend on a manually set fixed empirical threshold to select regions based on their semantic feature distance to the query text embedding. This traditional approach often lacks universal accuracy, leading to challenges in precisely identifying specific target areas. Instead, our method treats the feature selection process as a hyperplane division within the feature space, retaining only those features that are highly relevant to the query. We leverage off-the-shelf 2D Referring Expression Segmentation (RES) models to fine-tune the semantic-space hyperplane, enabling a more precise distinction between target regions and others. This fine-tuning substantially improves the accuracy of open-vocabulary queries, ensuring the precise localization of pertinent 3D Gaussians. Extensive experiments demonstrate GOI's superiority over previous state-of-the-art methods. Our project page is available at https://quyans.github.io/GOI-Hyperplane/ .

Multimodal Large Language Models (MLLMs) have shown significant potential for chart understanding and generation. However, they are still far from achieving the desired effectiveness in practical applications. This could be due to the limitations of the used training chart data. Existing chart datasets suffer from scarcity of chart types, limited coverage of tasks, and insufficient scalability, making them incapable of effectively enhancing the chart-related capabilities of MLLMs. To tackle these obstacles, we construct NovaChart, a large-scale dataset for chart understanding and generation of MLLMs. NovaChart contains 47K high-resolution chart images and 856K chart-related instructions, covering 18 different chart types and 15 unique tasks of chart understanding and generation. To build NovaChart, we propose a data generation engine for metadata curation, chart visualization and instruction formulation. Chart metadata in NovaChart contains detailed annotations, i.e., data points, visual elements, source data and the visualization code of every chart. This additional information endows NovaChart with considerable scalability, as it can facilitate the extension of chart instruction data to a larger scale and greater diversity. We utilize NovaChart to train several open-source MLLMs. Experimental results demonstrate NovaChart empowers MLLMs with stronger capabilities in 15 chart understanding and generation tasks by a large-margin (35.47%-619.47%), bringing them a step closer to smart chart assistants. Our dataset is now available at https://github.com/Elucidator-V/NovaChart.

In this paper, we explore the use of large language models (LLMs) to enhance video moment retrieval (VMR) by integrating general knowledge and pseudo-events as priors. We address the limitations of LLMs in generating continuous outputs, such as salience scores and inter-frame embeddings, which are critical for capturing inter-frame relations. To address these limitations, we propose using LLM encoders, which refine inter-concept relations in multimodal embeddings effectively, even without textual training. Our feasibility study shows that this capability extends to other embeddings like BLIP and T5 when they exhibit similar patterns to CLIP embeddings. We present a general framework for integrating LLM encoders into existing VMR architectures, specifically within the fusion module. The LLM encoder's ability to refine concept relation can help the model to achieve a balanced understanding of the foreground concepts (e.g., persons, faces) and background concepts (e.g., street, mountains) rather focusing only on the visually dominant foreground concepts. Additionally, we utilize pseudo-events, identified via event detection, to guide accurate moment prediction within event boundaries, reducing distractions from adjacent moments. Our plug-in approach for semantic refinement and pseudo-event regulation demonstrates state-of-the-art VMR performance through experimental validation. The source code can be accessed at https://github.com/fletcherjiang/LLMEPET.

Estimating the 3D poses of interacting hands from a monocular image is challenging due to the similarity in appearance between hand parts. Therefore, utilizing the appearance features alone tends to result in unreliable pose estimation. Existing approaches directly fuse the appearance features with position features, ignoring that the two types of features are heterogeneous. Here, the appearance features are derived from the RGB values of pixels, while the position features are mapped from the coordinates of pixels or joints. To address this problem, we present a novel framework called Decoupled Feature Learning (DFL ) for 3D pose estimation of interacting hands. By decoupling the appearance and position features, we facilitate the interactions within each feature type and those between both types of features. First, we compute the appearance relationships between the joint queries and the image feature maps; we utilize these relationships to aggregate each joint's appearance and position features. Second, we compute the 3D spatial relationships between hand joints using their position features; we utilize these relationships to guide the feature enhancement of joints. Third, we calculate appearance relationships and spatial relationships between the joints and image using the appearance and position features, respectively; we utilize these complementary relationships to promote the joints' location in the image. The two processes mentioned above are conducted iteratively. Finally, only the refined position features are used for hand pose estimation. This strategy avoids the step of mapping heterogeneous appearance features to hand-joint positions. Our method significantly outperforms state-of-the-art methods on the large-scale InterHand2.6M dataset. More impressively, our method exhibits strong generalization ability on in-the-wild images.

In the field of artificial intelligence, AI models are frequently described as 'black boxes' due to the obscurity of their internal mechanisms. It has ignited research interest on model interpretability, especially in attribution methods that offers precise explanations of model decisions. Current attribution algorithms typically evaluate the importance of each parameter by exploring the sample space. A large number of intermediate states are introduced during the exploration process, which may reach the model's Out-of-Distribution (OOD) space. Such intermediate states will impact the attribution results, making it challenging to grasp the relative importance of features. In this paper, we firstly define the local space and its relevant properties, and we propose the Local Attribution (LA) algorithm that leverages these properties. The LA algorithm comprises both targeted and untargeted exploration phases, which are designed to effectively generate intermediate states for attribution that thoroughly encompass the local space. Compared to the state-of-the-art attribution methods, our approach achieves an average improvement of 38.21% in attribution effectiveness. Extensive ablation studies in our experiments also validate the significance of each component in our algorithm. Our code is available at: https://github.com/LMBTough/LA/

Referring video object segmentation (RVOS) is a cross-modal task that aims to segment the target object described by language expressions. A video typically consists of multiple frames and existing works conduct segmentation at either the clip-level or the frame-level. Clip-level methods process a clip at once and segment in parallel, lacking explicit inter-frame interactions. In contrast, frame-level methods facilitate direct interactions between frames by processing videos frame by frame, but they are prone to error accumulation. In this paper, we propose a novel tracking-forced framework, introducing high-quality tracking information and forcing the model to achieve accurate segmentation. Concretely, we utilize the ground-truth segmentation of previous frames as accurate inter-frame interactions, providing high-quality tracking references for segmentation in the next frame. This decouples the current input from the previous output, which enables our model to concentrate on accurately segmenting just based on given tracking information, improving training efficiency and preventing error accumulation. For the inference stage without ground-truth masks, we carefully select the beginning frame to construct tracking information, aiming to ensure accurate tracking-based frame-by-frame object segmentation. With these designs, our tracking-forced method significantly outperforms existing methods on 4 widely used benchmarks by at least 3%. Especially, our method achieves 88.3% [email protected] accuracy and 87.6 overall IoU score on the JHMDB-Sentences dataset, surpassing previous best methods by 5.0% and 8.0, respectively.

Explaining what part of the input images primarily contributed to the predicted classification results by deep models has been widely researched over the years and many effective methods have been reported in the literature, for which deep Taylor decomposition (DTD) served as the primary foundation due to its advantage in theoretical explanations brought in by Taylor expansion and approximation. Recent research, however, has shown that the root of Taylor decomposition could extend beyond local linearity, and thus causing DTD to fail in delivering expected performances. In this paper, we propose a universal root inference method to overcome the shortfall and strengthen the roles of DTD in explainability and interpretability of deep classifications. In comparison with the existing approaches, our proposed features in: (i) theoretical establishment of the relationship between ideal roots and the propagated relevances; (ii) exploitation of gradient descents in learning a universal root inference; and (iii) constrained optimization of its final root selection. Extensive experiments, including both quantitative and qualitative, validate that our proposed root inference is not only effective, but also delivers significantly improved performances in explaining a range of deep classifiers. We share our codes via the link: https://github.com/meetxinzhang/XAI-RootInference.

Few-shot semantic segmentation (FSS) aims to locate pixels of unseen classes with clues from a few labeled samples. Recently, thanks to profound prior knowledge, diffusion models have been expanded to achieve FSS tasks. However, due to probabilistic noising and denoising processes, it is difficult for them to maintain spatial relationships between inputs and outputs, leading to inaccurate segmentation masks. To address this issue, we propose a Diffusion-based Segmentation network (DiffSeg), which decouples probabilistic denoising and segmentation processes. Specifically, DiffSeg leverages attention maps extracted from a pretrained diffusion model as support-query interaction information to guide segmentation, which mitigates the impact of probabilistic processes while benefiting from rich prior knowledge of diffusion models. In the segmentation stage, we present a Perceptual Attention Module (PAM), where two cross-attention mechanisms capture semantic information of support-query interaction and spatial information produced by the pretrained diffusion model. Furthermore, a self-attention mechanism within PAM ensures a balanced dependence for segmentation, thus preventing inconsistencies between the aforementioned semantic and spatial information. Additionally, considering the uncertainty inherent in the generation process of diffusion models, we equip DiffSeg with a Spatial Control Module (SCM), which models spatial structural information of query images to control boundaries of attention maps, thus aligning the spatial location between knowledge representation and query images. Experiments on PASCAL-5i and COCO datasets show that DiffSeg achieves new state-of-the-art performance with remarkable advantages.

ISP (Image Signal Processor) serves as a pipeline converting unprocessed raw images to sRGB images, positioned before nearly all visual tasks. Due to the varying spectral sensitivities of cameras, raw images captured by different cameras exist in different color spaces, making it challenging to deploy ISP across cameras with consistent performance. To address this challenge, it is intuitively to incorporate a raw-to-raw mapping (mapping raw images across camera color spaces) module into the ISP. However, the lack of paired data (i.e., images of the same scene captured by different cameras) makes it difficult to train a raw-to-raw model using supervised learning methods. In this paper, we aim to achieve ISP generalization by proposing the first unsupervised raw-to-raw model. To be specific, we propose a CSTPP (Color Space Transformation Parameters Predictor) module to predict the space transformation parameters in a patch-wise manner, which can accurately perform color space transformation and flexibly manage complex lighting conditions. Additionally, we design a CycleGAN-style training framework to realize unsupervised learning, overcoming the deficiency of paired data. Our proposed unsupervised model achieved performance comparable to that of the state-of-the-art semi-supervised method in raw-to-raw task. Furthermore, to assess its ability to generalize the ISP model across different cameras, we for the first formulated cross-camera ISP task and demonstrated the performance of our method through extensive experiments. The codes are released at https://github.com/ydxxxx/Unsupervised-Raw-to-raw-Mapping.

Image generation models can generate or edit images from a given text. Recent advancements in image generation technology, exemplified by DALL-E and Midjourney, have been groundbreaking. These advanced models, despite their impressive capabilities, are often trained on massive Internet datasets, making them susceptible to generating content that perpetuates social stereotypes and biases, which can lead to severe consequences. Prior research on assessing bias within image generation models suffers from several shortcomings, including limited accuracy, reliance on extensive human labor, and lack of comprehensive analysis. In this paper, we propose BiasPainter, a novel evaluation framework that can accurately, automatically and comprehensively trigger social bias in image generation models. BiasPainter uses a diverse range of seed images of individuals and prompts the image generation models to edit these images using gender, race, and age-neutral queries. These queries span 62 professions, 39 activities, 57 types of objects, and 70 personality traits. The framework then compares the edited images to the original seed images, focusing on the significant changes related to gender, race, and age. BiasPainter adopts a key insight that these characteristics should not be modified when subjected to neutral prompts. Built upon this design, BiasPainter can trigger the social bias and evaluate the fairness of image generation models. We use BiasPainter to evaluate six widely-used image generation models, such as stable diffusion and Midjourney. Experimental results show that BiasPainter can successfully trigger social bias in image generation models. According to our human evaluation, BiasPainter can achieve 90.8% accuracy on automatic bias detection, which is significantly higher than the results reported in previous work.

Multi-planar magnetic resonance imaging (MRI) can provide comprehensive 3D structural information for disease diagnosis. Compared to multi-source MRI, multi-planar MRI scans target areas in the human body from different directions. This atypical difference between directions may lead to poor performance of traditional domain generalization methods, especially when MRI from different planes also comes from different sources. In this paper, we propose ADDG, an Adaptive Domain Generalization framework for accurate cross-plane MRI segmentation. ADDG significantly mitigates the impact of information loss caused by slice spacing by injecting 3D shape prior to the segmentation target and capturing domain-agnostic feature differences from heterogeneous data sources through an adaptive data partitioning strategy. In addition, we propose a mesh deformation-based organ segmentation network to simultaneously delineate 2D boundary and 3D volume of organ, which could guide more accurate mesh deformation. We also develop an organ-specific mesh template and employ Loop subdivision for generating smoother 3D organ mesh. Furthermore, we design a flexible meta-learning paradigm to adaptively partition data domains based on invariant learning, which can learn domain-agnostic features from multi-source data to enhance the overall generalization ability. Experimental results show that ADDG outperforms several medical image segmentation, single-view 3D shape reconstruction, and domain generalization methods.

The rapid evolution of multimedia and computer vision technologies requires adaptive visual model deployment strategies to effectively handle diverse tasks and varying environments. This work introduces AxiomVision, a novel framework that can guarantee accuracy by leveraging edge computing to dynamically select the most efficient visual models for video analytics under diverse scenarios. Utilizing a tiered edge-cloud architecture, AxiomVision enables the deployment of a broad spectrum of visual models, from lightweight to complex DNNs, that can be tailored to specific scenarios while considering camera source impacts. In addition, AxiomVision provides three core innovations: (1) a dynamic visual model selection mechanism utilizing continual online learning, (2) an efficient online method that efficiently takes into account the influence of the camera's perspective, and (3) a topology-driven grouping approach that accelerates the model selection process. With rigorous theoretical guarantees, these advancements provide a scalable and effective solution for visual tasks inherent to multimedia systems, such as object detection, classification, and counting. Empirically, AxiomVision achieves a 25.7% improvement in accuracy.

3D visual grounding is a fundamental yet important task in multimedia understanding, which aims to locate a specific object in a complicated 3D scene semantically according to a text description. However, this task requires a large number of annotations of labeled text-object pairs for training, so the scarcity of annotated data has been a key obstacle in this task. To this end, this paper makes the first attempt to introduce and address a new semi-supervised setting, where only a few text-object labels are provided during training. Considering most scene data has no annotation, we explore a new solution for unlabeled 3D grounding by additionally training and transferring knowledge from a correlated task, i.e., 3D captioning. Our main insight is that 3D grounding and captioning are complementary and can be iteratively trained with unlabeled data to provide object and text contexts for each other with pseudo-label learning. Specifically, we propose a novel 3D Cross-Task Teacher-Student Framework (3D-CTTSF) for joint 3D grounding and captioning in the semi-supervised setting, where each branch contains parallel grounding and captioning modules. We first pre-train the two modules of the teacher branch with limited labeled data for warm-up. Then, we train the student branch to mimic the ability of the teacher model and iteratively update both branches with the unlabeled data. In particular, we transfer the learned knowledge between the grounding and captioning modules across two branches to generate and refine the pseudo-labels of unlabeled data for providing reliable supervision. To further improve the quality of the pseudo-labels, we design a cross-task pseudo-label generation scheme, filtering low-quality pseudo-labels at the detection, captioning, and grounding levels, respectively. Experimental results on various datasets show competitive performances in both tasks compared to previous fully- and weakly-supervised methods, demonstrating the proposed 3D-CTTSF can serve as an effective solution to overcome the data scarcity issue.

Multi-view learning methods often focus on improving decision accuracy, while neglecting the decision uncertainty, limiting their suitability for safety-critical applications. To mitigate this, researchers propose trusted multi-view learning methods that estimate classification probabilities and uncertainty by learning the class distributions for each instance. However, these methods assume that the data from each view can effectively differentiate all categories, ignoring the semantic vagueness phenomenon in real-world multi-view data. Our findings demonstrate that this phenomenon significantly suppresses the learning of view-specific evidence in existing methods. We propose a Consistent and Complementary-aware trusted Multi-view Learning (CCML) method to solve this problem. We first construct view opinions using evidential deep neural networks, which consist of belief mass vectors and uncertainty estimates. Next, we dynamically decouple the consistent and complementary evidence. The consistent evidence is derived from the shared portions across all views, while the complementary evidence is obtained by averaging the differing portions across all views. We ensure that the opinion constructed from the consistent evidence strictly aligns with the ground-truth category. For the opinion constructed from the complementary evidence, we allow it for potential vagueness in the evidence. We compare CCML with state-of-the-art baselines on one synthetic and six real-world datasets. The results validate the effectiveness of the dynamic evidence decoupling strategy and show that CCML significantly outperforms baselines on accuracy and reliability. The code is released at https://github.com/Lihong-Liu/CCML.

Long-tailed recognition (LTR) aims to learn balanced models from extremely unbalanced training data. Fine-tuning pretrained foundation models has recently emerged as a promising research direction for LTR. However, we observe that the fine-tuning process tends to degrade the intrinsic representation capability of pretrained models and lead to model bias towards certain classes, thereby hindering the overall recognition performance. To unleash the intrinsic representation capability of pretrained foundation models, in this work, we propose a new Parameter-Efficient Complementary Expert Learning (PECEL) for LTR. Specifically, PECEL consists of multiple experts, where individual experts are trained via Parameter-Efficient Fine-Tuning (PEFT) and encouraged to learn different expertise on complementary sub-categories via the proposed sample-aware logit adjustment loss. By aggregating the predictions of different experts, PECEL effectively achieves a balanced performance on long-tailed classes. Nevertheless, learning multiple experts generally introduces extra trainable parameters. To ensure parameter efficiency, we further propose a parameter sharing strategy which decomposes and shares the parameters in each expert. Extensive experiments on 4 LTR benchmarks show that the proposed PECEL can effectively learn multiple complementary experts without increasing the trainable parameters and achieve new state-of-the-art performance.

Lane detection (LD) is an essential component of autonomous driving systems, providing fundamental functionalities like adaptive cruise control and automated lane centering. Existing LD benchmarks primarily focus on evaluating common cases, neglecting the robustness of LD models against environmental illusions such as shadows and tire marks on the road. This research gap poses significant safety challenges since these illusions exist naturally in real-world traffic situations. For the first time, this paper studies the potential threats caused by these environmental illusions to LD and establishes the first comprehensive benchmark LanEvil for evaluating the robustness of LD against this natural corruption. We systematically design 14 prevalent yet critical types of environmental illusions (e.g., shadow, reflection) that cover a wide spectrum of real-world influencing factors in LD tasks. Based on real-world environments, we create 94 realistic and customizable 3D cases using the widely used CARLA simulator, resulting in a dataset comprising 90,292 sampled images. Through extensive experiments, we benchmark the robustness of popular LD methods using LanEvil, revealing substantial performance degradation (-5.37% Accuracy and -10.70% F1-Score on average), with shadow effects posing the greatest risk (-7.39% Accuracy). Additionally, we assess the performance of commercial auto-driving systems OpenPilot and Apollo through collaborative simulations, demonstrating that proposed environmental illusions can lead to incorrect decisions and potential traffic accidents. To defend against environmental illusions, we propose the Attention Area Mixing (AAM) approach using hard examples, which witness significant robustness improvement (+3.76%) under illumination effects. We hope our paper can contribute to advancing more robust auto-driving systems in the future. Part of our dataset and demos can be found at the https://lanevil.github.io/.