“CVPR 2025 將于 6 月 11 日至 15 日在美國(guó)納什維爾舉行。MMLab @HKU 攜24 篇高質(zhì)量論文隆重亮相?！?/strong>

01

CVPR 2025 納什維爾開(kāi)幕在即：

MMLab 攜前沿成果深度參與

作為計(jì)算機(jī)視覺(jué)領(lǐng)域最具影響力的國(guó)際會(huì)議之一，CVPR（IEEE Conference on Computer Vision and Pattern Recognition）每年都匯聚了全球頂尖高校、研究機(jī)構(gòu)與產(chǎn)業(yè)界的最新突破與前沿成果。CVPR 2025 將于6 月 11 日至 15 日在美國(guó)納什維爾舉行。MMLab@HKU 攜24 篇高質(zhì)量論文隆重亮相，涵蓋圖像生成、視頻理解、具身智能、三維重建、多模態(tài)融合等多個(gè)研究熱點(diǎn)。歡迎大家前來(lái)與論文作者面對(duì)面交流！

相關(guān)網(wǎng)站：https://mmlab.hk/event/cvpr2025

02

三大國(guó)際競(jìng)賽：以賽促研，智造未來(lái)

在 CVPR 2025，MMLab是三項(xiàng)國(guó)際競(jìng)賽的發(fā)起與主辦方，涵蓋開(kāi)放世界自動(dòng)駕駛以及機(jī)器人交互智能等多個(gè)熱門(mén)方向。通過(guò)精心設(shè)計(jì)的任務(wù)設(shè)置與評(píng)測(cè)機(jī)制，團(tuán)隊(duì)為全球研究者搭建了一個(gè)聚焦真實(shí)場(chǎng)景挑戰(zhàn)與技術(shù)落地能力的競(jìng)技舞臺(tái)。我們希望以此激發(fā)更多創(chuàng)新靈感，共同拓展視覺(jué)智能的未來(lái)邊界。

• Autonomous Grand Challenge 2025

• End-to-End Autonomous Driving through V2X Cooperation

• RoboTwin Dual-Arm Collaboration Challenge

03

六場(chǎng)深度活動(dòng)：解鎖 AI 落地的技術(shù)密碼

除了國(guó)際競(jìng)賽，MMLab 在 CVPR 2025 也主辦了六項(xiàng)前沿 Workshop、Tutorial 活動(dòng)，全面覆蓋自動(dòng)駕駛、多模態(tài)、世界模型、協(xié)同感知、數(shù)據(jù)賦能等熱點(diǎn)議題。

• Embodied Intelligence for Autonomous Systems on the Horizon

• Workshop on Autonomous Driving

• Distillation of Foundation Models for Autonomous Driving

• Multi-Agent Embodied Intelligent Systems Meet Generative-AI Era: Opportunities, Challenges and Futures

• Robotics 101: An Odyssey from A Vision Perspective

• The 1st Workshop on Benchmarking World Models

04

技術(shù)風(fēng)向標(biāo)：多項(xiàng)AI前沿研究盤(pán)點(diǎn)

在生成式智能與多模態(tài)感知飛速發(fā)展的當(dāng)下，這一系列研究成果展示了在跨模態(tài)理解、場(chǎng)景生成、人機(jī)交互和機(jī)器人智能等領(lǐng)域的一些進(jìn)步。比如，文本驅(qū)動(dòng)的視頻合成、圖像安全性評(píng)估、高精度的三維高斯建模和機(jī)器人操作策略學(xué)習(xí)這些技術(shù)，都在提升模型的通用性、效率以及在現(xiàn)實(shí)世界中的適應(yīng)能力。不管你關(guān)心的是更安全可信的生成系統(tǒng)、更聰明的機(jī)器人大腦，還是更高質(zhì)量的視覺(jué)生成模型，這些項(xiàng)目都代表了技術(shù)創(chuàng)新的前沿，歡迎關(guān)注！

• TokenHSI: Unified Synthesis of Physical Human-Scene Interactions through Task Tokenization[Oral]

• • 統(tǒng)一物理人景交互合成，通過(guò)任務(wù)分詞實(shí)現(xiàn)

  • arXiv: https://arxiv.org/abs/2503.19901

  • Github: https://github.com/liangpan99/TokenHSI

• Parallelized Autoregressive Visual Generation[Highlight]

• • PAR，根據(jù)視覺(jué)token間依賴(lài)關(guān)系所設(shè)計(jì)的并行自回歸生成模型

  • arXiv: https://arxiv.org/abs/2412.15119

  • Github: https://yuqingwang1029.github.io/PAR-project/

• RoboTwin: Dual-Arm Robot Benchmark with Generative Digital Twins[Highlight]

• • 機(jī)器人雙臂評(píng)測(cè)基準(zhǔn)集與數(shù)據(jù)合成器

  • arXiv: https://arxiv.org/abs/2504.13059

  • Github: https://github.com/TianxingChen/RoboTwin

• HMAR: Efficient Hierarchical Masked AutoRegressive Image Generation

• • HMAR，通過(guò)多尺度自回歸與掩碼重建結(jié)合的高效高質(zhì)量圖像生成模型

  • arXiv: https://arxiv.org/html/2506.04421v1

  • Project Page: https://research.nvidia.com/labs/dir/hmar/

• MBQ: Modality-Balanced Quantization for Large Vision-Language Models

• • MBQ，均衡視覺(jué)和語(yǔ)言之間敏感性差異的視覺(jué)-語(yǔ)言模型量化方法

  • arXiv: https://arxiv.org/abs/2412.19509

  • Github: https://github.com/thu-nics/MBQ

• MIDI: Multi-Instance Diffusion for Single Image to 3D Scene Generation

• • MIDI-3D，拓展3D物體生成模型到可組合的3D場(chǎng)景生成。

  • Arxiv: https://arxiv.org/abs/2412.03558

  • Github: https://github.com/VAST-AI-Research/MIDI-3D

• T2ISafety: Benchmark for Assessing Fairness, Toxicity, and Privacy in Image Generation

• • T2ISafety，一個(gè)評(píng)估文生圖模型安全性的基準(zhǔn)

  • Arxiv: https://arxiv.org/abs/2501.12612

  • Github: https://github.com/adwardlee/t2i_safety

• T2V-CompBench: A Comprehensive Benchmark for Compositional Text-to-video Generation

• • T2V-CompBench，評(píng)估文生視頻模型的組合生成能力

  • arXiv: https://arxiv.org/abs/2407.14505

  • Github: https://t2v-compbench-2025.github.io/

• CompGS: Unleashing 2D Compositionality for Compositional Text-to-3D via Dynamically Optimizing 3D Gaussians

• • 通過(guò) 3D Gaussian Splatting 實(shí)現(xiàn)高效的組合文本到三維內(nèi)容生成

  • arXiv: https://arxiv.org/abs/2410.20723

  • Github: https://chongjiange.github.io/compgs.html

• DRiVE: Diffusion-based Rigging Empowers Generation of Versatile and Expressive Characters

• • 擴(kuò)散模型驅(qū)動(dòng)，生成多變3D角色并自動(dòng)綁定

  • arXiv: https://arxiv.org/abs/2411.17423

  • Github: https://github.com/yisuanwang/DRiVE

• DexHandDiff: Interaction-aware Diffusion Planning for Adaptive Dexterous Manipulation

• • 自適應(yīng)靈巧手操作的交互感知擴(kuò)散策略

  • arXiv: https://arxiv.org/abs/2411.18562

  • Github: https://dexdiffuser.github.io/

• Distilling Monocular Foundation Model for Fine-grained Depth Completion

• • 知識(shí)蒸餾得到單目基礎(chǔ)模型用于將稀疏深度稠密化

  • arXiv: https://arxiv.org/abs/2503.16970

  • Github: https://github.com/Sharpiless/DMD3C

• Flashgs: Efficient 3d gaussian splatting for large-scale and high-resolution rendering

• • 高效 3D Gaussian Splatting，實(shí)現(xiàn)大規(guī)模高分辨率渲染

  • arXiv: https://arxiv.org/abs/2408.07967

  • Github: https://github.com/InternLandMark/FlashGS

• Forensics-Bench: A Comprehensive Forgery Detection Benchmark Suite for Large Vision Language Models

• • 針對(duì)大型視覺(jué)語(yǔ)言模型偽造檢測(cè)能力的全新綜合評(píng)測(cè)基準(zhǔn)

  • arXiv: https://arxiv.org/pdf/2503.15024

  • Github: https://github.com/Forensics-Bench/Forensics-Bench

  • project page: https://forensics-bench.github.io/

  • dataset:https://huggingface.co/datasets/Forensics-bench/Forensics-bench

• G3Flow: Generative 3D Semantic Flow for Pose-aware and Generalizable Object Manipulation

• • 生成式機(jī)器人3D操作增強(qiáng)表征

  • arXiv: https://arxiv.org/abs/2411.18369

  • Github: https://github.com/TianxingChen/G3Flow

• GraphMimic: Graph-to-Graphs Generative Modeling from Videos for Policy Learning

• • 通過(guò)視頻預(yù)訓(xùn)練，根據(jù)當(dāng)前圖結(jié)構(gòu)預(yù)測(cè)生成未來(lái)圖結(jié)構(gòu)

• Horizon-GS: Unified 3D Gaussian Splatting for Large-Scale Aerial-to-Ground Scenes

• • 統(tǒng)一空地視角3D Gaussian Splatting，重建渲染大場(chǎng)景

  • arXiv: https://arxiv.org/abs/2412.01745

  • Github: https://github.com/OpenRobotLab/HorizonGS

• Janus: Decoupling visual encoding for unified multimodal understanding and generation

• • 解耦視覺(jué)編碼以實(shí)現(xiàn)統(tǒng)一的多模態(tài)理解和生成

  • arXiv: https://arxiv.org/abs/2410.13848

  • Github: https://github.com/deepseek-ai/Janus

• JiSAM: Alleviate Labeling Burden and Corner Case Problems in Autonomous Driving via Minimal Real-World Data

• • 用CARLA的仿真數(shù)據(jù)幫助減少自動(dòng)駕駛感知中實(shí)際數(shù)據(jù)標(biāo)注的需求

  • arXiv: https://arxiv.org/abs/2503.08422

  • Github: https://github.com/Runjian-Chen/JiSAM

• MangaNinja: Line Art Colorization with Precise Reference Following

• • 精準(zhǔn)可控的線(xiàn)稿上色

  • arXiv: https://arxiv.org/abs/2501.08332

  • Github: https://github.com/ali-vilab/MangaNinjia

• NADER: Neural Architecture Design via Multi-Agent Collaboration

• • 多智能體協(xié)作的神經(jīng)網(wǎng)絡(luò)架構(gòu)設(shè)計(jì)

  • arXiv: https://arxiv.org/abs/2412.19206

• OpenING: A Comprehensive Benchmark for Judging Open-ended Interleaved Image-Text Generation

• • 首個(gè)面向開(kāi)放式圖文交錯(cuò)生成任務(wù)的綜合評(píng)測(cè)基準(zhǔn)

  • arXiv: https://arxiv.org/abs/2411.18499

  • Github: https://opening-benchmark.github.io/

• RoboBrain: A Unified Brain Model for Robotic Manipulation from Abstract to Concrete

• • 面向機(jī)器人操控的 MLLM 大模型

  • arXiv: https://arxiv.org/abs/2502.21257

  • Github: https://github.com/FlagOpen/RoboBrain

• ScaMo: Exploring the Scaling Law in Autoregressive Motion Generation Model

• • 探索自回歸運(yùn)動(dòng)生成模型的縮放定律

  • arXiv: https://arxiv.org/abs/2412.14559

  • Github: https://github.com/shunlinlu/ScaMo_code

• EdgeTAM: On-Device Track Anything Model

• • 壓縮視頻分割基模型SAM2，保持模型效果同時(shí)實(shí)現(xiàn)端側(cè)部署

  • arXiv: https://arxiv.org/abs/2501.07256

未經(jīng)「AI科技評(píng)論」授權(quán)，嚴(yán)禁以任何方式在網(wǎng)頁(yè)、論壇、社區(qū)進(jìn)行轉(zhuǎn)載！

公眾號(hào)轉(zhuǎn)載請(qǐng)先在「AI科技評(píng)論」后臺(tái)留言取得授權(quán)，轉(zhuǎn)載時(shí)需標(biāo)注來(lái)源并插入本公眾號(hào)名片。