Visual SLAMMOT Considering Multiple Motion Models

Simultaneous Localization and Mapping (SLAM) and Multi-Object Tracking (MOT) are pivotal tasks in the realm of autonomous driving, attracting considerable research attention. While SLAM endeavors to generate real-time maps and determine the vehicle's pose in unfamiliar settings, MOT focuses on the real-time identification and tracking of multiple dynamic objects. Despite their importance, the prevalent approach treats SLAM and MOT as independent modules within an autonomous vehicle system, leading to inherent limitations. Classical SLAM methodologies often rely on a static environment assumption, suitable for indoor rather than dynamic outdoor scenarios. Conversely, conventional MOT techniques typically rely on the vehicle's known state, constraining the accuracy of object state estimations based on this prior. To address these challenges, previous efforts introduced the unified SLAMMOT paradigm, yet primarily focused on simplistic motion patterns. In our team's previous work IMM-SLAMMOT\cite{IMM-SLAMMOT}, we present a novel methodology incorporating consideration of multiple motion models into SLAMMOT i.e. tightly coupled SLAM and MOT, demonstrating its efficacy in LiDAR-based systems. This paper studies feasibility and advantages of instantiating this methodology as visual SLAMMOT, bridging the gap between LiDAR and vision-based sensing mechanisms. Specifically, we propose a solution of visual SLAMMOT considering multiple motion models and validate the inherent advantages of IMM-SLAMMOT in the visual domain.

Communication-Efficient Cooperative SLAMMOT via Determining the Number of Collaboration Vehicles

The SLAMMOT, i.e. simultaneous localization, mapping, and moving object (detection and) tracking, represents an emerging technology for autonomous vehicles in dynamic environments. Such single-vehicle systems still have inherent limitations, such as occlusion issues. Inspired by SLAMMOT and rapidly evolving cooperative technologies, it is natural to explore cooperative simultaneous localization, mapping, moving object (detection and) tracking (C-SLAMMOT) to enhance state estimation for ego-vehicles and moving objects. C-SLAMMOT could significantly upgrade the single-vehicle performance by utilizing and integrating the shared information through communication among the multiple vehicles. This inevitably leads to a fundamental trade-off between performance and communication cost, especially in a scalable manner as the number of collaboration vehicles increases. To address this challenge, we propose a LiDAR-based communication-efficient C-SLAMMOT (CE C-SLAMMOT) method by determining the number of collaboration vehicles. In CE C-SLAMMOT, we adopt descriptor-based methods for enhancing ego-vehicle pose estimation and spatial confidence map-based methods for cooperative object perception, allowing for the continuous and dynamic selection of the corresponding critical collaboration vehicles and interaction content. This approach avoids the waste of precious communication costs by preventing the sharing of information from certain collaborative vehicles that may contribute little or no performance gain, compared to the baseline method of exchanging raw observation information among all vehicles. Comparative experiments in various aspects have confirmed that the proposed method achieves a good trade-off between performance and communication costs, while also outperforms previous state-of-the-art methods in cooperative perception performance.

LLaVA-CoT: Let Vision Language Models Reason Step-by-Step

Large language models have demonstrated substantial advancements in reasoning capabilities, particularly through inference-time scaling, as illustrated by models such as OpenAI's o1. However, current Vision-Language Models (VLMs) often struggle to perform systematic and structured reasoning, especially when handling complex visual question-answering tasks. In this work, we introduce LLaVA-CoT, a novel VLM designed to conduct autonomous multistage reasoning. Unlike chain-of-thought prompting, LLaVA-CoT independently engages in sequential stages of summarization, visual interpretation, logical reasoning, and conclusion generation. This structured approach enables LLaVA-CoT to achieve marked improvements in precision on reasoning-intensive tasks. To accomplish this, we compile the LLaVA-CoT-100k dataset, integrating samples from various visual question answering sources and providing structured reasoning annotations. Besides, we propose an inference-time stage-level beam search method, which enables effective inference-time scaling. Remarkably, with only 100k training samples and a simple yet effective inference time scaling method, LLaVA-CoT not only outperforms its base model by 8.9% on a wide range of multimodal reasoning benchmarks, but also surpasses the performance of larger and even closed-source models, such as Gemini-1.5-pro, GPT-4o-mini, and Llama-3.2-90B-Vision-Instruct.

DGTR: Distributed Gaussian Turbo-Reconstruction for Sparse-View Vast Scenes

Novel-view synthesis (NVS) approaches play a critical role in vast scene reconstruction. However, these methods rely heavily on dense image inputs and prolonged training times, making them unsuitable where computational resources are limited. Additionally, few-shot methods often struggle with poor reconstruction quality in vast environments. This paper presents DGTR, a novel distributed framework for efficient Gaussian reconstruction for sparse-view vast scenes. Our approach divides the scene into regions, processed independently by drones with sparse image inputs. Using a feed-forward Gaussian model, we predict high-quality Gaussian primitives, followed by a global alignment algorithm to ensure geometric consistency. Synthetic views and depth priors are incorporated to further enhance training, while a distillation-based model aggregation mechanism enables efficient reconstruction. Our method achieves high-quality large-scale scene reconstruction and novel-view synthesis in significantly reduced training times, outperforming existing approaches in both speed and scalability. We demonstrate the effectiveness of our framework on vast aerial scenes, achieving high-quality results within minutes. Code will released on our [https://3d-aigc.github.io/DGTR].

Motion Control for Enhanced Complex Action Video Generation

Existing text-to-video (T2V) models often struggle with generating videos with sufficiently pronounced or complex actions. A key limitation lies in the text prompt's inability to precisely convey intricate motion details. To address this, we propose a novel framework, MVideo, designed to produce long-duration videos with precise, fluid actions. MVideo overcomes the limitations of text prompts by incorporating mask sequences as an additional motion condition input, providing a clearer, more accurate representation of intended actions. Leveraging foundational vision models such as GroundingDINO and SAM2, MVideo automatically generates mask sequences, enhancing both efficiency and robustness. Our results demonstrate that, after training, MVideo effectively aligns text prompts with motion conditions to produce videos that simultaneously meet both criteria. This dual control mechanism allows for more dynamic video generation by enabling alterations to either the text prompt or motion condition independently, or both in tandem. Furthermore, MVideo supports motion condition editing and composition, facilitating the generation of videos with more complex actions. MVideo thus advances T2V motion generation, setting a strong benchmark for improved action depiction in current video diffusion models. Our project page is available at https://mvideo-v1.github.io/.

Exploring Hierarchical Molecular Graph Representation in Multimodal LLMs

Following the milestones in large language models (LLMs) and multimodal models, we have seen a surge in applying LLMs to biochemical tasks. Leveraging graph features and molecular text representations, LLMs can tackle various tasks, such as predicting chemical reaction outcomes and describing molecular properties. However, most current work overlooks the multi-level nature of graph features. The impact of different feature levels on LLMs and the importance of each level remain unexplored, and it is possible that different chemistry tasks require different feature levels. In this work, we first investigate the effect of feature granularity by fusing GNN-generated feature tokens, discovering that even reducing all tokens to a single token does not significantly impact performance. We then explore the effect of various feature levels on performance, finding that both the quality of LLM-generated molecules and performance on different tasks benefit from different feature levels. We conclude with two key insights: (1) current molecular Multimodal LLMs(MLLMs) lack a comprehensive understanding of graph features, and (2) static processing is not sufficient for hierarchical graph feature. Our code will be publicly available soon.

InjecGuard: Benchmarking and Mitigating Over-defense in Prompt Injection Guardrail Models

Prompt injection attacks pose a critical threat to large language models (LLMs), enabling goal hijacking and data leakage. Prompt guard models, though effective in defense, suffer from over-defense -- falsely flagging benign inputs as malicious due to trigger word bias. To address this issue, we introduce NotInject, an evaluation dataset that systematically measures over-defense across various prompt guard models. NotInject contains 339 benign samples enriched with trigger words common in prompt injection attacks, enabling fine-grained evaluation. Our results show that state-of-the-art models suffer from over-defense issues, with accuracy dropping close to random guessing levels (60%). To mitigate this, we propose InjecGuard, a novel prompt guard model that incorporates a new training strategy, Mitigating Over-defense for Free (MOF), which significantly reduces the bias on trigger words. InjecGuard demonstrates state-of-the-art performance on diverse benchmarks including NotInject, surpassing the existing best model by 30.8%, offering a robust and open-source solution for detecting prompt injection attacks. The code and datasets are released at https://github.com/SaFoLab-WISC/InjecGuard.

Fidelity-Imposed Displacement Editing for the Learn2Reg 2024 SHG-BF Challenge

Co-examination of second-harmonic generation (SHG) and bright-field (BF) microscopy enables the differentiation of tissue components and collagen fibers, aiding the analysis of human breast and pancreatic cancer tissues. However, large discrepancies between SHG and BF images pose challenges for current learning-based registration models in aligning SHG to BF. In this paper, we propose a novel multi-modal registration framework that employs fidelity-imposed displacement editing to address these challenges. The framework integrates batch-wise contrastive learning, feature-based pre-alignment, and instance-level optimization. Experimental results from the Learn2Reg COMULISglobe SHG-BF Challenge validate the effectiveness of our method, securing the 1st place on the online leaderboard.

Generative Adversarial Patches for Physical Attacks on Cross-Modal Pedestrian Re-Identification

Visible-infrared pedestrian Re-identification (VI-ReID) aims to match pedestrian images captured by infrared cameras and visible cameras. However, VI-ReID, like other traditional cross-modal image matching tasks, poses significant challenges due to its human-centered nature. This is evidenced by the shortcomings of existing methods, which struggle to extract common features across modalities, while losing valuable information when bridging the gap between them in the implicit feature space, potentially compromising security. To address this vulnerability, this paper introduces the first physical adversarial attack against VI-ReID models. Our method, termed Edge-Attack, specifically tests the models' ability to leverage deep-level implicit features by focusing on edge information, the most salient explicit feature differentiating individuals across modalities. Edge-Attack utilizes a novel two-step approach. First, a multi-level edge feature extractor is trained in a self-supervised manner to capture discriminative edge representations for each individual. Second, a generative model based on Vision Transformer Generative Adversarial Networks (ViTGAN) is employed to generate adversarial patches conditioned on the extracted edge features. By applying these patches to pedestrian clothing, we create realistic, physically-realizable adversarial samples. This black-box, self-supervised approach ensures the generalizability of our attack against various VI-ReID models. Extensive experiments on SYSU-MM01 and RegDB datasets, including real-world deployments, demonstrate the effectiveness of Edge- Attack in significantly degrading the performance of state-of-the-art VI-ReID methods.

STTATTS: Unified Speech-To-Text And Text-To-Speech Model

Speech recognition and speech synthesis models are typically trained separately, each with its own set of learning objectives, training data, and model parameters, resulting in two distinct large networks. We propose a parameter-efficient approach to learning ASR and TTS jointly via a multi-task learning objective and shared parameters. Our evaluation demonstrates that the performance of our multi-task model is comparable to that of individually trained models while significantly saving computational and memory costs ($\sim$50\% reduction in the total number of parameters required for the two tasks combined). We experiment with English as a resource-rich language, and Arabic as a relatively low-resource language due to shortage of TTS data. Our models are trained with publicly available data, and both the training code and model checkpoints are openly available for further research.