Memory Storyboard: Leveraging Temporal Segmentation for Streaming Self-Supervised Learning from Egocentric Videos

Self-supervised learning holds the promise to learn good representations from real-world continuous uncurated data streams. However, most existing works in visual self-supervised learning focus on static images or artificial data streams. Towards exploring a more realistic learning substrate, we investigate streaming self-supervised learning from long-form real-world egocentric video streams. Inspired by the event segmentation mechanism in human perception and memory, we propose "Memory Storyboard" that groups recent past frames into temporal segments for more effective summarization of the past visual streams for memory replay. To accommodate efficient temporal segmentation, we propose a two-tier memory hierarchy: the recent past is stored in a short-term memory, and the storyboard temporal segments are then transferred to a long-term memory. Experiments on real-world egocentric video datasets including SAYCam and KrishnaCam show that contrastive learning objectives on top of storyboard frames result in semantically meaningful representations which outperform those produced by state-of-the-art unsupervised continual learning methods.

Scaling up self-supervised learning for improved surgical foundation models

Foundation models have revolutionized computer vision by achieving vastly superior performance across diverse tasks through large-scale pretraining on extensive datasets. However, their application in surgical computer vision has been limited. This study addresses this gap by introducing SurgeNetXL, a novel surgical foundation model that sets a new benchmark in surgical computer vision. Trained on the largest reported surgical dataset to date, comprising over 4.7 million video frames, SurgeNetXL achieves consistent top-tier performance across six datasets spanning four surgical procedures and three tasks, including semantic segmentation, phase recognition, and critical view of safety (CVS) classification. Compared with the best-performing surgical foundation models, SurgeNetXL shows mean improvements of 2.4, 9.0, and 12.6 percent for semantic segmentation, phase recognition, and CVS classification, respectively. Additionally, SurgeNetXL outperforms the best-performing ImageNet-based variants by 14.4, 4.0, and 1.6 percent in the respective tasks. In addition to advancing model performance, this study provides key insights into scaling pretraining datasets, extending training durations, and optimizing model architectures specifically for surgical computer vision. These findings pave the way for improved generalizability and robustness in data-scarce scenarios, offering a comprehensive framework for future research in this domain. All models and a subset of the SurgeNetXL dataset, including over 2 million video frames, are publicly available at: https://github.com/TimJaspers0801/SurgeNet.

Few-shot Structure-Informed Machinery Part Segmentation with Foundation Models and Graph Neural Networks

This paper proposes a novel approach to few-shot semantic segmentation for machinery with multiple parts that exhibit spatial and hierarchical relationships. Our method integrates the foundation models CLIPSeg and Segment Anything Model (SAM) with the interest point detector SuperPoint and a graph convolutional network (GCN) to accurately segment machinery parts. By providing 1 to 25 annotated samples, our model, evaluated on a purely synthetic dataset depicting a truck-mounted loading crane, achieves effective segmentation across various levels of detail. Training times are kept under five minutes on consumer GPUs. The model demonstrates robust generalization to real data, achieving a qualitative synthetic-to-real generalization with a $J\&F$ score of 92.2 on real data using 10 synthetic support samples. When benchmarked on the DAVIS 2017 dataset, it achieves a $J\&F$ score of 71.5 in semi-supervised video segmentation with three support samples. This method's fast training times and effective generalization to real data make it a valuable tool for autonomous systems interacting with machinery and infrastructure, and illustrate the potential of combined and orchestrated foundation models for few-shot segmentation tasks.

Aligning First, Then Fusing: A Novel Weakly Supervised Multimodal Violence Detection Method

Weakly supervised violence detection refers to the technique of training models to identify violent segments in videos using only video-level labels. Among these approaches, multimodal violence detection, which integrates modalities such as audio and optical flow, holds great potential. Existing methods in this domain primarily focus on designing multimodal fusion models to address modality discrepancies. In contrast, we take a different approach; leveraging the inherent discrepancies across modalities in violence event representation to propose a novel multimodal semantic feature alignment method. This method sparsely maps the semantic features of local, transient, and less informative modalities ( such as audio and optical flow ) into the more informative RGB semantic feature space. Through an iterative process, the method identifies the suitable no-zero feature matching subspace and aligns the modality-specific event representations based on this subspace, enabling the full exploitation of information from all modalities during the subsequent modality fusion stage. Building on this, we design a new weakly supervised violence detection framework that consists of unimodal multiple-instance learning for extracting unimodal semantic features, multimodal alignment, multimodal fusion, and final detection. Experimental results on benchmark datasets demonstrate the effectiveness of our method, achieving an average precision (AP) of 86.07% on the XD-Violence dataset. Our code is available at https://github.com/xjpp2016/MAVD.

AVS-Mamba: Exploring Temporal and Multi-modal Mamba for Audio-Visual Segmentation

The essence of audio-visual segmentation (AVS) lies in locating and delineating sound-emitting objects within a video stream. While Transformer-based methods have shown promise, their handling of long-range dependencies struggles due to quadratic computational costs, presenting a bottleneck in complex scenarios. To overcome this limitation and facilitate complex multi-modal comprehension with linear complexity, we introduce AVS-Mamba, a selective state space model to address the AVS task. Our framework incorporates two key components for video understanding and cross-modal learning: Temporal Mamba Block for sequential video processing and Vision-to-Audio Fusion Block for advanced audio-vision integration. Building on this, we develop the Multi-scale Temporal Encoder, aimed at enhancing the learning of visual features across scales, facilitating the perception of intra- and inter-frame information. To perform multi-modal fusion, we propose the Modality Aggregation Decoder, leveraging the Vision-to-Audio Fusion Block to integrate visual features into audio features across both frame and temporal levels. Further, we adopt the Contextual Integration Pyramid to perform audio-to-vision spatial-temporal context collaboration. Through these innovative contributions, our approach achieves new state-of-the-art results on the AVSBench-object and AVSBench-semantic datasets. Our source code and model weights are available at AVS-Mamba.

OCORD: Open-Campus Object Removal Dataset

The rapid advancements in generative models, particularly diffusion-based techniques, have revolutionized image inpainting tasks by enabling the generation of high-fidelity and diverse content. However, object removal remains under-explored as a specific subset of inpainting, facing challenges such as inadequate semantic understanding and the unintended generation of artifacts. Existing datasets for object removal often rely on synthetic data, which fails to align with real-world scenarios, limiting model performance. Although some real-world datasets address these issues partially, they suffer from scalability, annotation inefficiencies, and limited realism in physical phenomena such as lighting and shadows. To address these limitations, this paper introduces a novel approach to object removal by constructing a high-resolution real-world dataset through long-duration video capture with fixed camera settings. Leveraging advanced tools such as Grounding-DINO, Segment-Anything-Model, and MASA for automated annotation, we provides image, background, and mask pairs while significantly reducing annotation time and labor. With our efficient annotation pipeline, we release the first fully open, high-resolution real-world dataset for object removal, and improved performance in object removal tasks through fine-tuning of pre-trained diffusion models.

Multilevel Semantic-Aware Model for AI-Generated Video Quality Assessment

The rapid development of diffusion models has greatly advanced AI-generated videos in terms of length and consistency recently, yet assessing AI-generated videos still remains challenging. Previous approaches have often focused on User-Generated Content(UGC), but few have targeted AI-Generated Video Quality Assessment methods. In this work, we introduce MSA-VQA, a Multilevel Semantic-Aware Model for AI-Generated Video Quality Assessment, which leverages CLIP-based semantic supervision and cross-attention mechanisms. Our hierarchical framework analyzes video content at three levels: frame, segment, and video. We propose a Prompt Semantic Supervision Module using text encoder of CLIP to ensure semantic consistency between videos and conditional prompts. Additionally, we propose the Semantic Mutation-aware Module to capture subtle variations between frames. Extensive experiments demonstrate our method achieves state-of-the-art results.

LiDAR-Camera Fusion for Video Panoptic Segmentation without Video Training

Panoptic segmentation, which combines instance and semantic segmentation, has gained a lot of attention in autonomous vehicles, due to its comprehensive representation of the scene. This task can be applied for cameras and LiDAR sensors, but there has been a limited focus on combining both sensors to enhance image panoptic segmentation (PS). Although previous research has acknowledged the benefit of 3D data on camera-based scene perception, no specific study has explored the influence of 3D data on image and video panoptic segmentation (VPS).This work seeks to introduce a feature fusion module that enhances PS and VPS by fusing LiDAR and image data for autonomous vehicles. We also illustrate that, in addition to this fusion, our proposed model, which utilizes two simple modifications, can further deliver even more high-quality VPS without being trained on video data. The results demonstrate a substantial improvement in both the image and video panoptic segmentation evaluation metrics by up to 5 points.

PanoSLAM: Panoptic 3D Scene Reconstruction via Gaussian SLAM

Understanding geometric, semantic, and instance information in 3D scenes from sequential video data is essential for applications in robotics and augmented reality. However, existing Simultaneous Localization and Mapping (SLAM) methods generally focus on either geometric or semantic reconstruction. In this paper, we introduce PanoSLAM, the first SLAM system to integrate geometric reconstruction, 3D semantic segmentation, and 3D instance segmentation within a unified framework. Our approach builds upon 3D Gaussian Splatting, modified with several critical components to enable efficient rendering of depth, color, semantic, and instance information from arbitrary viewpoints. To achieve panoptic 3D scene reconstruction from sequential RGB-D videos, we propose an online Spatial-Temporal Lifting (STL) module that transfers 2D panoptic predictions from vision models into 3D Gaussian representations. This STL module addresses the challenges of label noise and inconsistencies in 2D predictions by refining the pseudo labels across multi-view inputs, creating a coherent 3D representation that enhances segmentation accuracy. Our experiments show that PanoSLAM outperforms recent semantic SLAM methods in both mapping and tracking accuracy. For the first time, it achieves panoptic 3D reconstruction of open-world environments directly from the RGB-D video. (https://github.com/runnanchen/PanoSLAM)

PG-SAG: Parallel Gaussian Splatting for Fine-Grained Large-Scale Urban Buildings Reconstruction via Semantic-Aware Grouping

3D Gaussian Splatting (3DGS) has emerged as a transformative method in the field of real-time novel synthesis. Based on 3DGS, recent advancements cope with large-scale scenes via spatial-based partition strategy to reduce video memory and optimization time costs. In this work, we introduce a parallel Gaussian splatting method, termed PG-SAG, which fully exploits semantic cues for both partitioning and Gaussian kernel optimization, enabling fine-grained building surface reconstruction of large-scale urban areas without downsampling the original image resolution. First, the Cross-modal model - Language Segment Anything is leveraged to segment building masks. Then, the segmented building regions is grouped into sub-regions according to the visibility check across registered images. The Gaussian kernels for these sub-regions are optimized in parallel with masked pixels. In addition, the normal loss is re-formulated for the detected edges of masks to alleviate the ambiguities in normal vectors on edges. Finally, to improve the optimization of 3D Gaussians, we introduce a gradient-constrained balance-load loss that accounts for the complexity of the corresponding scenes, effectively minimizing the thread waiting time in the pixel-parallel rendering stage as well as the reconstruction lost. Extensive experiments are tested on various urban datasets, the results demonstrated the superior performance of our PG-SAG on building surface reconstruction, compared to several state-of-the-art 3DGS-based methods. Project Web:https://github.com/TFWang-9527/PG-SAG.