arXiv subjects

Shimin Chen

Publications and source records attributed to Shimin Chen.

18 recordsLinked to original sources

arXiv2026

RAIRS: Optimizing Redundant Assignment and List Layout for IVF-Based ANN Search

IVF is one of the most widely used ANNS (Approximate Nearest Neighbors Search) methods in vector databases. The idea of redundant assignment is to assign a data vector to more than one IVF lists for reducing the chance of missing true neighbors in IVF search. However, the naive strategy, which selects the second IVF list based on the distance between a data vector and the list centroids, performs poorly. Previous work focuses only on the inner product distance, while there is no optimized list selection study for the most popular Euclidean space. Moreover, the IVF search may access the same vector in more than one lists, resulting in redundant distance computation and decreasing query throughput. In this paper, we present RAIRS to address the above two challenges. For the challenge of the list selection, we propose an optimized AIR metric for the Euclidean space. AIR takes not only distances but also directions into consideration in order to support queries that are closer to the data vector but father away from the first chosen list's centroid. For the challenge of redundant distance computation, we propose SEIL, an optimized list layout that exploits shared cells to reduce repeated distance computations for IVF search. Our experimental results using representative real-world data sets show that RAIRS out-performs existing redundant assignment solutions and achieves up to 1.33x improvement over the best-performing IVF method, IVF-PQ Fast Scan with refinement.

cs.DB↗

arXiv2025

UniComp: Rethinking Video Compression Through Informational Uniqueness

Distinct from attention-based compression methods, this paper presents an information uniqueness driven video compression framework, termed UniComp, which aims to maximize the information fidelity of video representations under constrained computational budgets. Starting from the information-theoretic perspective, we formulate the vision compression as an optimization problem that minimizes conditional entropy (reconstruction error) between retained and full tokens. To achieve this, we introduce the notion of information uniqueness to measure intrinsic redundancy among tokens to link with reconstruction error. Based on uniqueness, we design three modules-Frame Group Fusion, Token Allocation, and Spatial Dynamic Compression-that progressively perform semantic frame grouping, adaptive resource allocation, and fine-grained spatial compression. Extensive experiments demonstrate that UniComp consistently outperforms existing compression methods in preserving essential visual tokens under limited computational budgets, highlighting the pivotal role of information uniqueness in token compression efficacy.

cs.CV↗

arXiv2025

LongCat-Flash-Omni Technical Report

We introduce LongCat-Flash-Omni, a state-of-the-art open-source omni-modal model with 560 billion parameters, excelling at real-time audio-visual interaction. By adopting a curriculum-inspired progressive training strategy that transitions from simpler to increasingly complex modality sequence modeling tasks, LongCat-Flash-Omni attains comprehensive multimodal capabilities while maintaining strong unimodal capability. Building upon LongCat-Flash, which adopts a high-performance Shortcut-connected Mixture-of-Experts (MoE) architecture with zero-computation experts, LongCat-Flash-Omni integrates efficient multimodal perception and speech reconstruction modules. Despite its immense size of 560B parameters (with 27B activated), LongCat-Flash-Omni achieves low-latency real-time audio-visual interaction. For training infrastructure, we developed a modality-decoupled parallelism scheme specifically designed to manage the data and model heterogeneity inherent in large-scale multimodal training. This innovative approach demonstrates exceptional efficiency by sustaining over 90% of the throughput achieved by text-only training. Extensive evaluations show that LongCat-Flash-Omni achieves state-of-the-art performance on omni-modal benchmarks among open-source models. Furthermore, it delivers highly competitive results across a wide range of modality-specific tasks, including text, image, and video understanding, as well as audio understanding and generation. We provide a comprehensive overview of the model architecture design, training procedures, and data strategies, and open-source the model to foster future research and development in the community.

cs.MM↗

arXiv2024

TimeMarker: A Versatile Video-LLM for Long and Short Video Understanding with Superior Temporal Localization Ability

Rapid development of large language models (LLMs) has significantly advanced multimodal large language models (LMMs), particularly in vision-language tasks. However, existing video-language models often overlook precise temporal localization and struggle with videos of varying lengths. We introduce TimeMarker, a versatile Video-LLM designed for high-quality dialogue based on video content, emphasizing temporal localization. TimeMarker integrates Temporal Separator Tokens to enhance temporal awareness, accurately marking specific moments within videos. It employs the AnyLength mechanism for dynamic frame sampling and adaptive token merging, enabling effective handling of both short and long videos. Additionally, TimeMarker utilizes diverse datasets, including further transformed temporal-related video QA datasets, to bolster its temporal understanding capabilities. Image and interleaved data are also employed to further enhance the model's semantic perception ability. Evaluations demonstrate that TimeMarker achieves state-of-the-art performance across multiple benchmarks, excelling in both short and long video categories. Our project page is at \url{https://github.com/TimeMarker-LLM/TimeMarker/}.

cs.CV↗

arXiv2024

Technical Report for SoccerNet Challenge 2022 -- Replay Grounding Task

In order to make full use of video information, we transform the replay grounding problem into a video action location problem. We apply a unified network Faster-TAD proposed by us for temporal action detection to get the results of replay grounding. Finally, by observing the data distribution of the training data, we refine the output of the model to get the final submission.

cs.CV↗

arXiv2024

Technical Report for Soccernet 2023 -- Dense Video Captioning

In the task of dense video captioning of Soccernet dataset, we propose to generate a video caption of each soccer action and locate the timestamp of the caption. Firstly, we apply Blip as our video caption framework to generate video captions. Then we locate the timestamp by using (1) multi-size sliding windows (2) temporal proposal generation and (3) proposal classification.

cs.CV↗

arXiv2024

Technical Report for ActivityNet Challenge 2022 -- Temporal Action Localization

In the task of temporal action localization of ActivityNet-1.3 datasets, we propose to locate the temporal boundaries of each action and predict action class in untrimmed videos. We first apply VideoSwinTransformer as feature extractor to extract different features. Then we apply a unified network following Faster-TAD to simultaneously obtain proposals and semantic labels. Last, we ensemble the results of different temporal action detection models which complement each other. Faster-TAD simplifies the pipeline of TAD and gets remarkable performance, obtaining comparable results as those of multi-step approaches.

cs.CV↗

arXiv2024

LITS: An Optimized Learned Index for Strings (An Extended Version)

Index is an important component in database systems. Learned indexes have been shown to outperform traditional tree-based index structures for fixed-sized integer or floating point keys. However, the application of the learned solution to variable-length string keys is under-researched. Our experiments show that existing learned indexes for strings fail to outperform traditional string indexes, such as HOT and ART. String keys are long and variable sized, and often contain skewed prefixes, which make the last-mile search expensive, and adversely impact the capability of learned models to capture the skewed distribution of string keys. In this paper, we propose a novel learned index for string keys, LITS (Learned Index with Hash-enhanced Prefix Table and Sub-tries). We propose an optimized learned model, combining a global Hash-enhanced Prefix Table (HPT) and a per-node local linear model to better distinguish string keys. Moreover, LITS exploits compact leaf nodes and hybrid structures with a PMSS model for efficient point and range operations. Our experimental results using eleven string data sets show that LITS achieves up to 2.43x and 2.27x improvement over HOT and ART for point operations, and attains comparable scan performance.

cs.DB↗

arXiv2024

Fewer Tokens and Fewer Videos: Extending Video Understanding Abilities in Large Vision-Language Models

Amidst the advancements in image-based Large Vision-Language Models (image-LVLM), the transition to video-based models (video-LVLM) is hindered by the limited availability of quality video data. This paper addresses the challenge by leveraging the visual commonalities between images and videos to efficiently evolve image-LVLMs into video-LVLMs. We present a cost-effective video-LVLM that enhances model architecture, introduces innovative training strategies, and identifies the most effective types of video instruction data. Our innovative weighted token sampler significantly compresses the visual token numbers of each video frame, effectively cutting computational expenses. We also find that judiciously using just 10% of the video data, compared to prior video-LVLMs, yields impressive results during various training phases. Moreover, we delve into the influence of video instruction data in limited-resource settings, highlighting the significance of incorporating video training data that emphasizes temporal understanding to enhance model performance. The resulting Fewer Tokens and Fewer Videos LVLM (FTFV-LVLM) exhibits exceptional performance across video and image benchmarks, validating our model's design and training approaches.

cs.CV↗

arXiv2023

SoccerNet 2023 Challenges Results

The SoccerNet 2023 challenges were the third annual video understanding challenges organized by the SoccerNet team. For this third edition, the challenges were composed of seven vision-based tasks split into three main themes. The first theme, broadcast video understanding, is composed of three high-level tasks related to describing events occurring in the video broadcasts: (1) action spotting, focusing on retrieving all timestamps related to global actions in soccer, (2) ball action spotting, focusing on retrieving all timestamps related to the soccer ball change of state, and (3) dense video captioning, focusing on describing the broadcast with natural language and anchored timestamps. The second theme, field understanding, relates to the single task of (4) camera calibration, focusing on retrieving the intrinsic and extrinsic camera parameters from images. The third and last theme, player understanding, is composed of three low-level tasks related to extracting information about the players: (5) re-identification, focusing on retrieving the same players across multiple views, (6) multiple object tracking, focusing on tracking players and the ball through unedited video streams, and (7) jersey number recognition, focusing on recognizing the jersey number of players from tracklets. Compared to the previous editions of the SoccerNet challenges, tasks (2-3-7) are novel, including new annotations and data, task (4) was enhanced with more data and annotations, and task (6) now focuses on end-to-end approaches. More information on the tasks, challenges, and leaderboards are available on https://www.soccer-net.org. Baselines and development kits can be found on https://github.com/SoccerNet.

cs.CV↗

arXiv2022

SoccerNet 2022 Challenges Results

The SoccerNet 2022 challenges were the second annual video understanding challenges organized by the SoccerNet team. In 2022, the challenges were composed of 6 vision-based tasks: (1) action spotting, focusing on retrieving action timestamps in long untrimmed videos, (2) replay grounding, focusing on retrieving the live moment of an action shown in a replay, (3) pitch localization, focusing on detecting line and goal part elements, (4) camera calibration, dedicated to retrieving the intrinsic and extrinsic camera parameters, (5) player re-identification, focusing on retrieving the same players across multiple views, and (6) multiple object tracking, focusing on tracking players and the ball through unedited video streams. Compared to last year's challenges, tasks (1-2) had their evaluation metrics redefined to consider tighter temporal accuracies, and tasks (3-6) were novel, including their underlying data and annotations. More information on the tasks, challenges and leaderboards are available on https://www.soccer-net.org. Baselines and development kits are available on https://github.com/SoccerNet.

cs.CV↗

arXiv2022

Density-optimized Intersection-free Mapping and Matrix Multiplication for Join-Project Operations (extended version)

A Join-Project operation is a join operation followed by a duplicate eliminating projection operation. It is used in a large variety of applications, including entity matching, set analytics, and graph analytics. Previous work proposes a hybrid design that exploits the classical solution (i.e., join and deduplication), and MM (matrix multiplication) to process the sparse and the dense portions of the input data, respectively. However, we observe three problems in the state-of-the-art solution: 1) The outputs of the sparse and dense portions overlap, requiring an extra deduplication step; 2) Its table-to-matrix transformation makes an over-simplified assumption of the attribute values; and 3) There is a mismatch between the employed MM in BLAS packages and the characteristics of the Join-Project operation. In this paper, we propose DIM3, an optimized algorithm for the Join-Project operation. To address 1), we propose an intersection-free partition method to completely remove the final deduplication step. For 2), we develop an optimized design for mapping attribute values to natural numbers. For 3), we propose DenseEC and SparseBMM algorithms to exploit the structure of Join-Project for better efficiency. Moreover, we extend DIM3 to consider partial result caching and support Join-op queries, including Join-Aggregate and MJP (Multi-way Joins with Projection). Experimental results using both real-world and synthetic data sets show that DIM3 outperforms previous Join-Project solutions by a factor of 2.3x-18x. Compared to RDBMSs, DIM3 achieves orders of magnitude speedups.

cs.DB↗

arXiv2022

Faster-TAD: Towards Temporal Action Detection with Proposal Generation and Classification in a Unified Network

Temporal action detection (TAD) aims to detect the semantic labels and boundaries of action instances in untrimmed videos. Current mainstream approaches are multi-step solutions, which fall short in efficiency and flexibility. In this paper, we propose a unified network for TAD, termed Faster-TAD, by re-purposing a Faster-RCNN like architecture. To tackle the unique difficulty in TAD, we make important improvements over the original framework. We propose a new Context-Adaptive Proposal Module and an innovative Fake-Proposal Generation Block. What's more, we use atomic action features to improve the performance. Faster-TAD simplifies the pipeline of TAD and gets remarkable performance on lots of benchmarks, i.e., ActivityNet-1.3 (40.01% mAP), HACS Segments (38.39% mAP), SoccerNet-Action Spotting (54.09% mAP). It outperforms existing single-network detector by a large margin.

cs.CV↗

arXiv2022

MM-SEAL: A Large-scale Video Dataset of Multi-person Multi-grained Spatio-temporally Action Localization

In this paper, we introduce a novel large-scale video dataset dubbed MM-SEAL for multi-person multi-grained spatio-temporal action localization among human daily life. We are the first to propose a new benchmark for multi-person spatio-temporal complex activity localization, where complex semantic and long duration bring new challenges to localization tasks. We observe that limited atomic actions can be combined into many complex activities. MM-SEAL provides both atomic action and complex activity annotations, producing 111.7k atomic actions spanning 172 action categories and 17.7k complex activities spanning 200 activity categories. We explore the relationship between atomic actions and complex activities, finding that atomic action features can improve the complex activity localization performance. Also, we propose a new network which generates temporal proposals and labels simultaneously, termed Faster-TAD. Finally, our evaluations show that visual features pretrained on MM-SEAL can improve the performance on other action localization benchmarks. We will release the dataset and the project code upon publication of the paper.

cs.CV↗

arXiv2021

Updatable Learned Index with Precise Positions

Index plays an essential role in modern database engines to accelerate the query processing. The new paradigm of "learned index" has significantly changed the way of designing index structures in DBMS. The key insight is that indexes could be regarded as learned models that predict the position of a lookup key in the dataset. While such studies show promising results in both lookup time and index size, they cannot efficiently support update operations. Although recent studies have proposed some preliminary approaches to support update, they are at the cost of scarifying the lookup performance as they suffer from the overheads brought by imprecise predictions in the leaf nodes. In this paper, we propose LIPP, a brand new framework of learned index to address such issues. Similar with state-of-the-art learned index structures, LIPP is able to support all kinds of index operations, namely lookup query, range query, insert, delete, update and bulkload. Meanwhile, we overcome the limitations of previous studies by properly extending the tree structure when dealing with update operations so as to eliminate the deviation of location predicted by the models in the leaf nodes. Moreover, we further propose a dynamic adjustment strategy to ensure that the height of the tree index is tightly bounded and provide comprehensive theoretical analysis to illustrate it. We conduct an extensive set of experiments on several real-life and synthetic datasets. The results demonstrate that our method consistently outperforms state-of-the-art solutions, achieving by up to 4x for a broader class of workloads with different index operations.

cs.DB↗

arXiv2020

Storing and Querying Large-Scale Spatio-Temporal Graphs with High-Throughput Edge Insertions

Real-world graphs often contain spatio-temporal information and evolve over time. Compared with static graphs, spatio-temporal graphs have very different characteristics, presenting more significant challenges in data volume, data velocity, and query processing. In this paper, we describe three representative applications to understand the features of spatio-temporal graphs. Based on the commonalities of the applications, we define a formal spatio-temporal graph model, where a graph consists of location vertices, object vertices, and event edges. Then we discuss a set of design goals to meet the requirements of the applications: (i) supporting up to 10 billion object vertices, 10 million location vertices, and 100 trillion edges in the graph, (ii) supporting up to 1 trillion new edges that are streamed in daily, and (iii) minimizing cross-machine communication for query processing. We propose and evaluate PAST, a framework for efficient PArtitioning and query processing of Spatio-Temporal graphs. Experimental results show that PAST successfully achieves the above goals. It improves query performance by orders of magnitude compared with state-of-the-art solutions, including JanusGraph, Greenplum, Spark and ST-Hadoop.

cs.DC↗

arXiv2018

Efficient Graph Computation for Node2Vec

Node2Vec is a state-of-the-art general-purpose feature learning method for network analysis. However, current solutions cannot run Node2Vec on large-scale graphs with billions of vertices and edges, which are common in real-world applications. The existing distributed Node2Vec on Spark incurs significant space and time overhead. It runs out of memory even for mid-sized graphs with millions of vertices. Moreover, it considers at most 30 edges for every vertex in generating random walks, causing poor result quality. In this paper, we propose Fast-Node2Vec, a family of efficient Node2Vec random walk algorithms on a Pregel-like graph computation framework. Fast-Node2Vec computes transition probabilities during random walks to reduce memory space consumption and computation overhead for large-scale graphs. The Pregel-like scheme avoids space and time overhead of Spark's read-only RDD structures and shuffle operations. Moreover, we propose a number of optimization techniques to further reduce the computation overhead for popular vertices with large degrees. Empirical evaluation show that Fast-Node2Vec is capable of computing Node2Vec on graphs with billions of vertices and edges on a mid-sized machine cluster. Compared to Spark-Node2Vec, Fast-Node2Vec achieves 7.7--122x speedups.

cs.DC↗

arXiv2015

i2MapReduce: Incremental MapReduce for Mining Evolving Big Data

As new data and updates are constantly arriving, the results of data mining applications become stale and obsolete over time. Incremental processing is a promising approach to refreshing mining results. It utilizes previously saved states to avoid the expense of re-computation from scratch. In this paper, we propose i2MapReduce, a novel incremental processing extension to MapReduce, the most widely used framework for mining big data. Compared with the state-of-the-art work on Incoop, i2MapReduce (i) performs key-value pair level incremental processing rather than task level re-computation, (ii) supports not only one-step computation but also more sophisticated iterative computation, which is widely used in data mining applications, and (iii) incorporates a set of novel techniques to reduce I/O overhead for accessing preserved fine-grain computation states. We evaluate i2MapReduce using a one-step algorithm and three iterative algorithms with diverse computation characteristics. Experimental results on Amazon EC2 show significant performance improvements of i2MapReduce compared to both plain and iterative MapReduce performing re-computation.

cs.DC↗