时长：
19分钟
播放：
114
发布：
4天前
主播...
简介...
你有没有想过，能写诗作画的AI，为什么有时却像个固执的孩子？本期我们要聊的几篇最新论文，就试图教会AI一些我们习以为常、但它却难以理解的人类智慧。我们将一起看看，如何治好AI的“路痴”症，让它拥有空间感；如何让它从被动看图，变身主动破案的“侦探”；甚至，如何通过巧妙的“换个姿势”，让它终于听懂“不要”，并随心所欲地调整观察事物的“粒度”。
00:00:33 人工智能的“路痴”难题
00:05:24 AI侦探，如何给千米大桥做“体检”？
00:09:59 从“你猜”到“你定”：AI图像分割的新玩法
00:14:45 换个姿势，让AI听懂“不要”
本期介绍的几篇论文：
[CV] Scaling Spatial Intelligence with Multimodal Foundation Models
[SenseTime Research]
https://arxiv.org/abs/2511.13719
---
[CV] BridgeEQA: Virtual Embodied Agents for Real Bridge Inspections
[University of Houston]
https://arxiv.org/abs/2511.12676
---
[CV] UnSAMv2: Self-Supervised Learning Enables Segment Anything at Any Granularity
[UC Berkeley]
https://arxiv.org/abs/2511.13714
---
[CV] SpaceVLM: Sub-Space Modeling of Negation in Vision-Language Models
[MIT]
https://arxiv.org/abs/2511.12331
00:00:33 人工智能的“路痴”难题
00:05:24 AI侦探，如何给千米大桥做“体检”？
00:09:59 从“你猜”到“你定”：AI图像分割的新玩法
00:14:45 换个姿势，让AI听懂“不要”
本期介绍的几篇论文：
[CV] Scaling Spatial Intelligence with Multimodal Foundation Models
[SenseTime Research]
https://arxiv.org/abs/2511.13719
---
[CV] BridgeEQA: Virtual Embodied Agents for Real Bridge Inspections
[University of Houston]
https://arxiv.org/abs/2511.12676
---
[CV] UnSAMv2: Self-Supervised Learning Enables Segment Anything at Any Granularity
[UC Berkeley]
https://arxiv.org/abs/2511.13714
---
[CV] SpaceVLM: Sub-Space Modeling of Negation in Vision-Language Models
[MIT]
https://arxiv.org/abs/2511.12331
评价...
空空如也
小宇宙热门评论...
暂无小宇宙热门评论
