Partially Relevant Video Retrieval (PRVR) aims to retrieve untrimmed videos containing relevant moments for a given text query. This task is extremely challenging, as untrimmed videos often include numerous actions and objects unrelated to the query. However, existing methods usually struggle with fine-grained action-object modeling, limiting their retrieval performance. To tackle this challenge, we introduce Action-and-object Aware Alignment for Partially Relevant Video Retrieval (A3PRVR), a dual-branch framework designed to enhance retrieval by improving the modeling of action-object relationships. Specifically, we propose a Query-specific Deformable Temporal Attention (Q-DTA) module to effectively capture action-relevant object information in video features, while filtering out irrelevant content. Additionally, we propose an action-and-object aware alignment module to enable fine-grained textual understanding and video-text alignment. It uses action- and object-aware contrastive losses to enhance the model's sensitivity to action-object distinctions in the text query. Compared to state-of-the-art methods, A3PRVR achieves an average relative gain of 6.5% in SumR across the Charades-STA, ActivityNet-Caption, and TVR datasets.

Learning multimodal representation is a fundamental task that supports a wide range of applications such as visual-text retrieval. While pioneering approaches e.g., CLIP paves the way by learning separated encoders for different modalities, they struggle to model complex interactions between modalities, resulting in inferior vision and language representation. Recently, researchers have begun to leverage powerful Large Vision-Language Models (LVLMs) for unimodal or multimodal encoding, showing substantial improvement over separated encoder methods. However, we find that directly adapting LVLMs to embedding models suffers from insufficient visual representation and coarse multimodal alignment. To address these issues, we propose a simple yet effective Fine-grained Alignment Matters (FAM) method to achieve fine-grained vision-language embedding learning with LVLMs. First, to close the gap between the pure generation and multimodal embedding using LVLMs, we propose Multi-granularity Aligned Contrastive (MAC) to explicitly learn and align fine-grained modality representations at multiple granularity levels using image-text pairs. Second, to mitigate the insufficiency of visual representation during adapting LVLMs to downstream embedding tasks, we propose a Vision Embedding Inversion Training (VEIN) strategy to encourage the extracted embeddings to preserve fine-grained visual features. Extensive experiments demonstrate the effectiveness of our method, which achieves superior performance on various downstream multimodal datasets.

Transformer-based Large Language Models (LLMs) have exhibited remarkable success in extensive tasks primarily attributed to self-attention mechanism, which requires a token to consider all preceding tokens as its context to compute attention. However, when the context length L becomes very large (e.g., 128K), the amount of potentially redundant information in the context tends to increase. The redundant context not only hampers the modeling representation performance but also incurs unnecessary computational and storage overhead. In this paper, we propose a plug-and-play Core Context Aware (CCA) Attention for efficient long-context modeling, comprising two complementary modules: 1) Globality-aware pooling module groups input tokens and dynamically compresses each group into one core token based on their significance. In this way, our method automatically focuses and strengthens core context while diminishing redundancy during the learning process, leading to effective long-term dependency modeling. 2) Locality-preserving module incorporates neighboring tokens to preserve local context for detailed representation. Notably, our CCA-Attention is able to replace the self-attention module in existing LLMs with minimal fine-tuning cost. Extensive experimental results show the superiority of our method in both long-context modeling and computational efficiency over state-of-the-art methods.

Reconstructing deformable soft tissues from endoscopic videos is a critical yet challenging task. Leveraging depth priors, deformable implicit neural representations have seen significant advancements in this field. However, depth priors from pre-trained depth estimation models are often coarse, and inaccurate depth supervision can severely impair the performance of these neural networks. Moreover, existing methods overlook local similarities in input sequences, which restricts their effectiveness in capturing local details and tissue deformations. In this paper, we introduce UW-DNeRF, a novel approach utilizing neural radiance fields for high-quality reconstruction of deformable tissues. We propose an uncertainty-guided depth supervision strategy to mitigate the impact of inaccurate depth information. This strategy relaxes hard depth constraints and unlocks the potential of implicit neural representations. In addition, we design a local window-based information sharing scheme. This scheme employs local window and keyframe deformation networks to construct deformations with local awareness and enhances the model's ability to capture fine details. We demonstrate the superiority of our method over state-of-the-art approaches on synthetic and in vivo endoscopic datasets. Code is available at: https://github.com/IRMVLab/UW-DNeRF.

Building a self-model for robots, enabling them to simulate their physical selves and predict future states without direct interaction with the physical world, is crucial for robot motion planning and control. Existing self-modeling methods primarily focus on rigid robots and typically require significant time, effort, and resources to gather training data. In this study, we introduce SoftNeRF, a self-supervised visual self-model designed for soft robots. We use a hybrid neural shape representation based on the Signed Distance Function (SDF) to capture both the geometry and complex nonlinear motion of soft robots. By leveraging differentiable rendering, our method learns a self-model from readily available RGB images, similar to how humans understand their physical state through reflection. To improve training efficiency and model accuracy, we propose an error-guided adaptive sampling strategy. SoftNeRF can serve as a plug-in for various downstream tasks, even when trained with data unrelated to those tasks. We demonstrate SoftNeRF’s ability to support shape prediction and motion planning for robots in both simulated and real-world environments. Furthermore, SoftNeRF excels in detecting and recovering from damage, thereby enhancing machine resilience. Code is available at: https://github.com/irmvlab/soft-nerf.

Estimating camera motion and continuously reconstructing dense scenes in deformable environments presents a complex and open challenge. Many existing approaches tend to rely on assumptions about the scene’s topology or the nature of deformable motion. However, these assumptions do not hold true in medical endoscopy applications. To address these challenges, we introduce DDS-SLAM, a novel dense deformable semantic neural SLAM that achieves accurate camera tracking, continuous dense scene reconstruction, and high-quality image rendering in deformable scenes. First, we propose a novel hybrid neural scene representation method capable of capturing both natural and artificial deformations. Additionally, by leveraging the 2D semantic information of the scene, we introduce a semantic loss function based on semantic distance fields. This approach guides network optimization at a higher level, thereby enhancing system performance. Furthermore, we validate our method through a series of experiments conducted on several representative medical datasets, demonstrating its superiority over other state-of-the-art approaches. The code is available at: https://github.com/IRMVLab/DDS-SLAM.