Vision-and-language navigation (VLN) aims to develop agents capable of navigating in realistic environments. While recent cross-modal training approaches have significantly improved navigation performance in both indoor and outdoor scenarios, aerial navigation over real-world cities remains underexplored primarily due to limited datasets and the difficulty of integrating visual and geographic information. To fill this gap, we introduce CityNav, the first large-scale real-world dataset for aerial VLN. Our dataset consists of 32,637 human demonstration trajectories, each paired with a natural language description, covering 4.65 km2 across two real cities: Cambridge and Birmingham. In contrast to existing datasets composed of synthetic scenes such as AerialVLN, our dataset presents a unique challenge because agents must interpret spatial relationships between real-world landmarks and the navigation destination, making CityNav an essential benchmark for advancing aerial VLN. Furthermore, as an initial step toward addressing this challenge, we provide a methodology of creating geographic semantic maps that can be used as an auxiliary modality input during navigation. In our experiments, we compare performance of three representative aerial VLN agents (Seq2seq, CMA and AerialVLN models) and demonstrate that the semantic map representation significantly improves their navigation performance.
论文检索
输入标题、作者或关键词，从 1,526 篇学术成果中精准定位
Perception-as-Control: Fine-grained Controllable Image Animation with 3D-aware Motion Representation
PDF ↗Motion-controllable image animation is a fundamental task with a wide range of potential applications. Recent works have made progress in controlling camera or object motion via various motion representations, while they still struggle to support collaborative camera and object motion control with adaptive control granularity. To this end, we introduce 3D-aware motion representation and propose an image animation framework, called Perception-as-Control, to achieve fine-grained collaborative motion control. Specifically, we construct 3D-aware motion representation from a reference image, manipulate it based on interpreted user instructions, and perceive it from different viewpoints. In this way, camera and object motions are transformed into intuitive and consistent visual changes. Then, our framework leverages the perception results as motion control signals, enabling it to support various motion-related video synthesis tasks in a unified and flexible way. Experiments demonstrate the superiority of the proposed approach.
Large multimodal models (LMMs) have exhibited proficiencies across many visual tasks. Although numerous well-known benchmarks exist to evaluate model performance, they increasingly have insufficient headroom. As such, there is a pressing need for a new generation of benchmarks challenging enough for the next generation of LMMs. One area that LMMs show potential is graph analysis, specifically, the tasks an analyst might typically perform when interpreting figures such as estimating the mean, intercepts or correlations of functions and data series. In this work, we introduce GRAB, a graph analysis benchmark, fit for current and future frontier LMMs. Our benchmark is predominantly synthetic, ensuring high-quality, noise-free questions. GRAB is comprised of 3284 questions, covering five tasks and 23 graph properties. We evaluate 20 LMMs on GRAB, finding it to be a challenging benchmark, with the highest performing model attaining a score of just 21.0%. Finally, we conduct various ablations to investigate where the models succeed and struggle. We release GRAB and a lightweight GRAB-Lite to encourage progress in this important, growing domain.
Disentangled World Models: Learning to Transfer Semantic Knowledge from Distracting Videos for Reinforcement Learning
PDF ↗Training visual reinforcement learning (RL) in practical scenarios presents a significant challenge, i.e., RL agents suffer from low sample efficiency in environments with variations. While various approaches have attempted to alleviate this issue by disentangled representation learning, these methods usually start learning from scratch without prior knowledge of the world. This paper, in contrast, tries to learn and understand underlying semantic variations from distracting videos via offline-to-online latent distillation and flexible disentanglement constraints. To enable effective cross-domain semantic knowledge transfer, we introduce an interpretable model-based RL framework, dubbed Disentangled World Models (DisWM). Specifically, we pretrain the action-free video prediction model offline with disentanglement regularization to extract semantic knowledge from distracting videos. The disentanglement capability of the pretrained model is then transferred to the world model through latent distillation. For finetuning in the online environment, we exploit the knowledge from the pretrained model and introduce a disentanglement constraint to the world model. During the adaptation phase, the incorporation of actions and rewards from online environment interactions enriches the diversity of the data, which in turn strengthens the disentangled representation learning. Experimental results validate the superiority of our approach on various benchmarks.
We introduce a novel visual tokenization framework that embeds a provable PCA-like structure into the latent token space. While existing visual tokenizers primarily optimize for reconstruction fidelity, they often neglect the structural properties of the latent space--a critical factor for both interpretability and downstream tasks. Our method generates a 1D causal token sequence for images, where each successive token contributes non-overlapping information with mathematically guaranteed decreasing explained variance, analogous to principal component analysis. This structural constraint ensures the tokenizer extracts the most salient visual features first, with each subsequent token adding diminishing yet complementary information. Additionally, we identified and resolved a semantic-spectrum coupling effect that causes the unwanted entanglement of high-level semantic content and low-level spectral details in the tokens by leveraging a diffusion decoder. Experiments demonstrate that our approach achieves state-of-the-art reconstruction performance and enables better interpretability to align with the human vision system. Moreover, autoregressive models trained on our token sequences achieve performance comparable to current state-of-the-art methods while requiring fewer tokens for training and inference.
Neurons: Emulating the Human Visual Cortex Improves Fidelity and Interpretability in fMRI-to-Video Reconstruction
PDF ↗Decoding visual stimuli from neural activity is essential for understanding the human brain. While fMRI methods have successfully reconstructed static images, fMRI-tovideo reconstruction faces challenges due to the need for capturing spatiotemporal dynamics like motion and scene transitions. Recent approaches have improved semantic and perceptual alignment but struggle to integrate coarse fMRI data with detailed visual features. Inspired by the hierarchical organization of the visual system, we propose NEURONS, a novel framework that decouples learning into four correlated sub-tasks: key object segmentation, concept recognition, scene description, and blurry video reconstruction. This approach simulates the visual cortex's functional specialization, allowing the model to capture diverse video content. In the inference stage, NEURONS generates robust conditioning signals for a pre-trained textto-video diffusion model to reconstruct the videos. Extensive experiments demonstrate that NEURONS outperforms state-of-the-art baselines, achieving solid improvements in video consistency (26.6%) and semantic-level accuracy (19.1%). Notably, NEURONS shows a strong functional correlation with the visual cortex, highlighting its potential for brain-computer interfaces and clinical applications. Code and model weights are available at: https://github.com/xmed-lab/NEURONS.
TAB: Transformer Attention Bottlenecks enable User Intervention and Debugging in Vision-Language Models
PDF ↗Multi-head self-attention (MHSA) is a key component of Transformers, a widely popular architecture in both language and vision. Multiple heads intuitively enable different parallel processes over the same input. Yet, they also obscure the attribution of each input patch to the output of a model. We propose a novel 1-head Transformer Attention Bottleneck (TAB) layer, inserted after the traditional MHSA architecture, to serve as an attention bottleneck for interpretability and intervention. Unlike standard self-attention, TAB constrains the total attention over all patches to \in [0, 1]. That is, when the total attention is 0, no visual information is propagated further into the network, and the vision-language model (VLM) would default to a generic, image-independent response. To demonstrate the advantages of TAB, we train VLMs with TAB to perform image-difference captioning. Over three datasets, our models perform similarly to baseline VLMs in captioning but the bottleneck is superior in localizing changes and in identifying when no changes occur. TAB is the first architecture to enable users to debug by editing attention, which often produces expected outputs by VLMs.
Looking in the Mirror: A Faithful Counterfactual Explanation Method for Interpreting Deep Image Classification Models
PDF ↗Counterfactual explanations (CFE) for deep image classifiers aim to reveal how minimal input changes lead to different model decisions, providing critical insights for model interpretation and improvement. However, existing CFE methods often rely on additional image encoders and generative models to create plausible images, neglecting the classifier's own feature space and decision boundaries. As such, they do not explain the intrinsic feature space and decision boundaries learned by the classifier. To address this limitation, we propose Mirror-CFE, a novel method that generates faithful counterfactual explanations by operating directly in the classifier's feature space, treating decision boundaries as mirrors that "reflect" feature representations in the mirror. Mirror-CFE learns a mapping function from feature space to image space while preserving distance relationships, enabling smooth transitions between source images and their counterfactuals. Through extensive experiments on four image datasets, we demonstrate that Mirror-CFE achieves superior performance in validity while maintaining input resemblance compared to state-of-the-art explanation methods. Finally, mirror-CFE provides interpretable visualization of the classifier's decision process by generating step-wise transitions that reveal how features evolve as classification confidence changes.
Despite the success of deep learning in close-set 3D object detection, existing approaches struggle with zero-shot generalization to novel objects and camera configurations. We introduce DetAny3D, a promptable 3D detection foundation model capable of detecting any novel object under arbitrary camera configurations using only monocular inputs. Training a foundation model for 3D detection is fundamentally constrained by the limited availability of annotated 3D data, which motivates DetAny3D to leverage the rich prior knowledge embedded in extensively pre-trained 2D foundation models to compensate for this scarcity. To effectively transfer 2D knowledge to 3D, DetAny3D incorporates two core modules: the 2D Aggregator, which aligns features from different 2D foundation models, and the 3D Interpreter with Zero-Embedding Mapping, which stabilizes early training in 2D-to-3D knowledge transfer. Experimental results validate the strong generalization of our DetAny3D, which not only achieves state-of-the-art performance on unseen categories and novel camera configurations, but also surpasses most competitors on in-domain data. DetAny3D sheds light on the potential of the 3D foundation model for diverse applications in real-world scenarios, e.g., rare object detection in autonomous driving, and demonstrates promise for further exploration of 3D-centric tasks in open-world settings. More visualization results can be found at our code repository.
Recently, Multimodal Large Language Models (MLLMs) have emerged as a leading framework for enhancing the ability of Large Language Models (LLMs) to interpret non-linguistic modalities. Despite their impressive capabilities, the robustness of MLLMs under conditions where one or more modalities are missing remains largely unexplored. In this paper, we investigate the extent to which MLLMs can maintain performance when faced with missing modality inputs. Moreover, we propose a novel framework to mitigate the aforementioned issue called retrieval-augmented generation for missing modalities (MissRAG). It consists of a novel multimodal RAG technique alongside a tailored prompt engineering strategy designed to enhance model robustness by mitigating the impact of absent modalities while preventing the burden of additional instruction tuning. To demonstrate the effectiveness of our techniques, we conduct comprehensive evaluations across five diverse datasets, covering tasks such as audio-visual question answering, audio-visual captioning, and multimodal sentiment analysis. Our source code is available at https://github.com/aimagelab/MissRAG.
Multimodal LLMs (MLLMs) have reached remarkable levels of proficiency in understanding multimodal inputs. However, understanding and interpreting the behavior of such complex models is a challenging task, not to mention the dynamic shifts that may occur during fine-tuning, or due to covariate shift between datasets. In this work, we apply concept-level analysis towards MLLM understanding. More specifically, we propose to map hidden states to interpretable visual and textual concepts. This enables us to more efficiently compare certain semantic dynamics, such as the shift from an original and fine-tuned model, revealing concept alteration and potential biases that may occur during fine-tuning. We also demonstrate the use of shift vectors to capture these concepts changes. These shift vectors allow us to recover fine-tuned concepts by applying simple, computationally inexpensive additive concept shifts in the original model. Finally, our findings also have direct applications for MLLM steering, which can be used for model debiasing as well as enforcing safety in MLLM output. All in all, we propose a novel, training-free, ready-to-use framework for MLLM behavior interpretability and control. Our implementation is publicly available.
In recent years, Open-Vocabulary Semantic Segmentation (OVSS) has been largely advanced. However, existing methods mostly rely on a pre-trained vision-language model (e.g., CLIP) and require a predefined set of classes to guide the semantic segmentation process during the inference. This not only narrows the application scenario but also constrains comprehension within a finite vocabulary. To overcome this, we reformulate OVSS as a text generation task and propose the CLIP-adapted Region-to-Text Network (CRTNet) that achieves vocabulary-free OVSS by generating category names and descriptions upon segmentation masks. The training process consists of two steps to ensure an accurate and detailed interpretation of the masked regions: (i) the initial step adapts CLIP visual features to mask-level proposal features using binarized masks extracted by a trained mask extractor, and (ii) the subsequent step involves selecting and aggregating these features to become text-aware by integrating CLIP text embeddings, effectively aligning visual data with corresponding linguistic data to facilitate region-to-text learning. Furthermore, we introduce a series of parsing and filtering techniques to integrate multiple sources of training data to improve the generalization ability of our model. Experiments demonstrate that our model not only excels in OVSS but also exhibits scalability and can be adapted to various foundation models (e.g., SAM) without being retrained.
Remote sensing object detection has made significant progress, but most studies still focus on closed-set detection, limiting generalization across diverse datasets. Open-vocabulary object detection (OVD) provides a solution by leveraging multimodal associations between text prompts and visual features. However, existing OVD methods for remote sensing (RS) images are constrained by small-scale datasets and fail to address the unique challenges of remote sensing interpretation, include oriented object detection and the need for both high precision and real-time performance in diverse scenarios. To tackle these challenges, we propose OpenRSD, a universal open-prompt RS object detection framework. OpenRSD supports multimodal prompts and integrates multi-task detection heads to balance accuracy and real-time requirements. Additionally, we design a multi-stage training pipeline to enhance the generalization of model. Evaluated on seven public datasets, OpenRSD demonstrates superior performance in oriented and horizontal bounding box detection, with real-time inference capabilities suitable for large-scale RS image analysis. Compared to YOLO-World, OpenRSD exhibits an 8.7% higher average precision and achieves an inference speed of 20.8 FPS. Codes and models will be released.
Constructing Ophthalmic MLLM for Positioning-diagnosis Collaboration Through Clinical Cognitive Chain Reasoning
PDF ↗Multimodal large language models (MLLMs) demonstrate significant potential in the field of medical diagnosis. However, they face critical challenges in specialized domains such as ophthalmology, particularly the fragmentation of annotation granularity and inconsistencies in clinical reasoning logic, which hinder precise cross-modal understanding. This paper introduces FundusExpert, an ophthalmology-specific MLLM with integrated positioning-diagnosis reasoning capabilities, along with FundusGen, a dataset constructed through the intelligent Fundus-Engine system. Fundus-Engine automates localization and leverages MLLM-based semantic expansion to integrate global disease classification, local object detection, and fine-grained feature analysis within a single fundus image. Additionally, by constructing a clinically aligned cognitive chain, it guides the model to generate interpretable reasoning paths. FundusExpert, fine-tuned with instruction data from FundusGen, achieves the best performance in ophthalmic question-answering tasks, surpassing the average accuracy of the 40B MedRegA by 26.6%. It also excels in zero-shot report generation tasks, achieving a clinical consistency of 77.0%, significantly outperforming GPT-4o's 47.6%. Furthermore, we reveal a scaling law between data quality and model capability (L \propto N^ 0.068 ), demonstrating that the cognitive alignment annotations in FundusGen enhance data utilization efficiency. By integrating region-level localization with diagnostic reasoning chains, our work develops a scalable, clinically-aligned MLLM and explores a pathway toward bridging the visual-language gap in specific MLLMs. Our project can be found at https://github.com/MeteorElf/FundusExpert.
Multimodal LLMs are turning their focus to video benchmarks, however most video benchmarks only provide outcome supervision, with no intermediate or interpretable reasoning steps. This makes it challenging to assess if models are truly able to combine perceptual and temporal information to reason about videos, or simply get the correct answer by chance or by exploiting linguistic biases. To remedy this, we provide a new video reasoning dataset called MINERVA for modern multimodal models. Each question in the dataset comes with 5 answer choices, as well as detailed, hand-crafted reasoning traces. Our dataset is multimodal, diverse in terms of video domain and length, and consists of complex multi-step questions. Extensive benchmarking shows that our dataset provides a challenge for frontier open-source and proprietary models. We perform fine-grained error analysis to identify common failure modes across various models, and create a taxonomy of reasoning errors. We use this to explore both human and LLM-as-a-judge methods for scoring video reasoning traces, and find that failure modes are primarily related to temporal localization, followed by visual perception errors, as opposed to logical or completeness errors. The dataset, along with questions, answer candidates and reasoning traces is publicly available under https://github.com/google-deepmind/neptune?tab=readme-ov-file#minerva https://github.com/google-deepmind/neptune?tab=readme-ov-file\#minerva.
How well can Multimodal Large Language Models (MLLMs) understand composite images? Composite images (CIs) are synthetic visuals created by merging multiple visual elements, such as charts, posters, or screenshots, rather than being captured directly by a camera. While CIs are prevalent in real-world applications, recent MLLM developments have primarily focused on interpreting natural images (NIs). Our research reveals that current MLLMs face significant challenges in accurately understanding CIs, often struggling to extract information or perform complex reasoning based on these images. We find that existing training data for CIs are mostly formatted for question-answer tasks (e.g., in datasets like ChartQA and ScienceQA), while high-quality image-caption datasets, critical for robust vision-language alignment, are only available for NIs. To bridge this gap, we introduce Composite Captions (CompCap), a flexible framework that leverages Large Language Models (LLMs) and automation tools to synthesize CIs with accurate and detailed captions. Using CompCap, we curate CompCap-118K, a dataset containing 118K image-caption pairs across six CI types. We validate the effectiveness of CompCap-118K by supervised fine-tuning MLLMs of three sizes: xGen-MM-inst.-4B and LLaVA-NeXT-Vicuna-7B/13B. Empirical results show that CompCap-118K significantly enhances MLLMs' understanding of CIs, yielding average gains of 1.7%, 2.0%, and 2.9% across eleven benchmarks, respectively.
We study how rich visual semantic information is represented within various layers and denoising timesteps of different diffusion architectures. We uncover monosemantic interpretable features by leveraging k-sparse autoencoders (k-SAE). We substantiate our mechanistic interpretations via transfer learning using light-weight classifiers on off-the-shelf diffusion models' features. On 4 datasets, we demonstrate the effectiveness of diffusion features for representation learning. We provide an in-depth analysis of how different diffusion architectures, pre-training datasets, and language model conditioning impacts visual representation granularity, inductive biases, and transfer learning capabilities. Our work is a critical step towards deepening interpretability of black-box diffusion models. Code and visualizations available at: https://github.com/revelio-diffusion/revelio
Current large vision-language models (LVLMs) typically employ a connector module to link visual features with text embeddings of large language models (LLMs) and use end-to-end training to achieve multi-modal understanding in a unified process. Well alignment needs high-quality pre-training data and a carefully designed training process. Current LVLMs face challenges when addressing complex vision-language reasoning tasks, with their reasoning capabilities notably lagging behind those of LLMs. This paper proposes a paradigm shift: instead of training end-to-end vision-language reasoning models, we advocate for developing a decoupled reasoning framework based on existing visual interpretation specialists and text-based reasoning LLMs. Our approach leverages (1) a dedicated vision-language model to transform the visual content of images into textual descriptions and (2) an LLM to perform reasoning according to the visual-derived text and the original question. This method presents a cost-efficient solution for multi-modal model development by optimizing existing models to work collaboratively, avoiding end-to-end development of vision-language models from scratch. By transforming images into language model-compatible text representations, it facilitates future low-cost and flexible upgrades to upcoming powerful LLMs. We introduce an outcome-rewarded joint-tuning strategy to optimize the cooperation between the visual interpretation and linguistic reasoning model. Evaluation results on vision-language benchmarks demonstrate that the decoupled reasoning framework outperforms recent LVLMs. Our approach yields particularly significant performance gains on visually intensive geometric mathematics problems. The code is available: https://github.com/guozix/DVLR.
Optimal Transport for Brain-Image Alignment: Unveiling Redundancy and Synergy in Neural Information Processing
PDF ↗The design of artificial neural networks (ANNs) is inspired by the structure of the human brain, and in turn, ANNs offer a potential means to interpret and understand brain signals. Existing methods primarily align brain signals with stimulus signals using Mean Squared Error (MSE), which focuses only on local point-wise alignment and ignores global matching, leading to coarse interpretations and inaccuracies in brain signal decoding. In this paper, we address these issues through optimal transport (OT) and theoretically demonstrate why OT provides a more effective alignment strategy than MSE. Specifically, we construct a transport plan between brain voxel embeddings and image embeddings, enabling more precise matching. By controlling the amount of transport, we mitigate the influence of redundant information. We apply our alignment model directly to the Brain Captioning task by feeding brain signals into a large language model (LLM) instead of images. Our approach achieves state-of-the-art performance across ten evaluation metrics, surpassing the previous best method by an average of 6.11% in singlesubject training and 3.81% in cross-subject training. Additionally, we have uncovered several insightful conclusions that align with existing brain research. We unveil the redundancy and synergy of brain information processing through region masking and data dimensionality reduction visualization experiments. We believe our approach paves the way for a more precise understanding of brain signals in the future. The code is available at https://github.com/NKUShaw/OT-Alignment4brain-to-image
The rapid development of Large Multimodal Models (LMMs) for 2D images and videos has spurred efforts to adapt these models for interpreting 3D scenes. However, the absence of large-scale 3D vision-language datasets has posed a significant obstacle. To address this issue, typical approaches focus on injecting 3D awareness into 2D LMMs by designing 3D input-level scene representations. This work provides a new perspective. We introduce reconstructive visual instruction tuning with 3D-awareness (ROSS3D), which integrates 3D-aware visual supervision into the training procedure. Specifically, it incorporates cross-view and global-view reconstruction. The former requires reconstructing masked views by aggregating overlapping information from other views. The latter aims to aggregate information from all available views to recover Bird's-Eye-View images, contributing to a comprehensive overview of the entire scene. Empirically, ROSS3D achieves state-of-the-art performance across various 3D scene understanding benchmarks. More importantly, our semi-supervised experiments demonstrate significant potential in leveraging large amounts of unlabeled 3D vision-only data.