ECCV 2026 倒计时 12 天：LeCun 压轴、中国 71M 模型、CMU 一键修 3D，顶会三大爆点抢先看

TL;DR - ECCV 2026’s newly announced program highlights world models, autonomous driving, embodied perception, and efficient real-time generation. Early project previews include a 71M-parameter portrait-generation model and a general video model for correcting artifacts across multiple 3D representations.

• Keynotes feature Yann LeCun, Wayve chief scientist Jamie Shotton, and Kristen Grauman, spanning foundation models, driving world models, video, and embodied perception.
• StreamTalk uses one 71M-parameter model for real-time avatars, telepresence, and interactive NPCs, emphasizing lightweight deployment.
• CMU’s FixAnything applies a unified video model to repair rendering artifacts from 3D Gaussian splats, meshes, and sparse point clouds.
• The conference also includes workshops on safe world models, medical video, and related vision topics, plus a physical-AI robotics demonstration.