🐦 Twitter Post Details

Viewing enriched Twitter post

@ManlingLi_

Planning with the views: Can VLMs predict how each camera move changes the view, and plan many such moves ahead? We introduce ViewSuite with 6 DoF camera control and ~165K task instances, testing: Path-to-View View-to-Path Interactive View Planning A sharp Planning Gap emerges: + can roughly "track" how camera action changes views - cannot "compose" a plan towards a target view at all We then try to teach VLMs with Reinforcement Learning. - RL cannot teach VLMs such planning ability, only 2.5% success rate with Qwen2.5-VL-7B. + With View Graph Distillation (our RL-Graph-SFT framework), 2.5% → 47.8% Below, we answer these questions: Q1. What are the failure modes? Q2. How can we make RL work? Q3. What has the model learned? Can we open up the model to see before/after? Can such spatial priors transfer to other view related tasks? Led by @James_KKW, great to work with @LINJIEFUN @zhengyuan_yang @shiqi_chen17 @wzenus @drfeifei @jiajunwu_cs Leonidas Guibas, Lijuan Wang. A joint efforts with @StanfordAILab @StanfordSVL @MSFTResearch.

📊 Media Metadata

{
  "media": [
    {
      "url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2067697152722415623/media_0.mp4",
      "media_url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2067697152722415623/media_0.mp4",
      "type": "video",
      "filename": "media_0.mp4"
    }
  ],
  "processed_at": "2026-06-18T21:36:23.779743",
  "pipeline_version": "2.0"
}

🔧 Raw API Response

{
  "type": "tweet",
  "id": "2067697152722415623",
  "url": "https://x.com/ManlingLi_/status/2067697152722415623",
  "twitterUrl": "https://twitter.com/ManlingLi_/status/2067697152722415623",
  "text": "Planning with the views:\n\nCan VLMs predict how each camera move changes the view, and plan many such moves ahead?\n\nWe introduce ViewSuite with 6 DoF camera control and ~165K task instances, testing:\nPath-to-View\nView-to-Path\nInteractive View Planning\n\nA sharp Planning Gap emerges:\n+ can roughly \"track\" how camera action changes views\n- cannot \"compose\" a plan towards a target view at all\n\nWe then try to teach VLMs with Reinforcement Learning.  - RL cannot teach VLMs such planning ability, only 2.5% success rate with Qwen2.5-VL-7B.\n+ With View Graph Distillation (our RL-Graph-SFT framework), 2.5% → 47.8%\n\nBelow, we answer these questions: \nQ1. What are the failure modes?\nQ2. How can we make RL work? \nQ3. What has the model learned? Can we open up the model to see before/after? Can such spatial priors transfer to other view related tasks?\n\nLed by @James_KKW, great to work with @LINJIEFUN @zhengyuan_yang @shiqi_chen17 @wzenus @drfeifei @jiajunwu_cs Leonidas Guibas, Lijuan Wang.\n\nA joint efforts with @StanfordAILab @StanfordSVL @MSFTResearch.",
  "source": "Twitter for iPhone",
  "retweetCount": 15,
  "replyCount": 1,
  "likeCount": 35,
  "quoteCount": 1,
  "viewCount": 3630,
  "createdAt": "Thu Jun 18 19:53:11 +0000 2026",
  "lang": "en",
  "bookmarkCount": 10,
  "isReply": false,
  "inReplyToId": null,
  "conversationId": "2067697152722415623",
  "displayTextRange": [
    0,
    272
  ],
  "inReplyToUserId": null,
  "inReplyToUsername": null,
  "author": {
    "type": "user",
    "userName": "ManlingLi_",
    "url": "https://x.com/ManlingLi_",
    "twitterUrl": "https://twitter.com/ManlingLi_",
    "id": "930946520433623040",
    "name": "Manling Li",
    "isVerified": false,
    "isBlueVerified": true,
    "verifiedType": null,
    "profilePicture": "https://pbs.twimg.com/profile_images/1882681476199976960/MopnNPms_normal.jpg",
    "coverPicture": "",
    "description": "",
    "location": "Stanford, CA",
    "followers": 10991,
    "following": 918,
    "status": "",
    "canDm": true,
    "canMediaTag": true,
    "createdAt": "Wed Nov 15 23:52:01 +0000 2017",
    "entities": {
      "description": {
        "urls": []
      },
      "url": {}
    },
    "fastFollowersCount": 0,
    "favouritesCount": 1327,
    "hasCustomTimelines": true,
    "isTranslator": false,
    "mediaCount": 152,
    "statusesCount": 799,
    "withheldInCountries": [],
    "affiliatesHighlightedLabel": {},
    "possiblySensitive": false,
    "pinnedTweetIds": [
      "2067697152722415623"
    ],
    "profile_bio": {
      "description": "Assistant Professor@NU, Amazon Scholar, Postdoc@Stanford, PhD@UIUC #NLP #CV Language+Vision/EmbodiedAI, Reasoning, Planning, Compositionality, Trustworthiness",
      "entities": {
        "description": {
          "hashtags": [
            {
              "indices": [
                67,
                71
              ],
              "text": "NLP"
            },
            {
              "indices": [
                72,
                75
              ],
              "text": "CV"
            }
          ]
        },
        "url": {
          "urls": [
            {
              "display_url": "limanling.github.io",
              "expanded_url": "https://limanling.github.io/",
              "indices": [
                0,
                23
              ],
              "url": "https://t.co/qfi75HiXXw"
            }
          ]
        }
      }
    },
    "isAutomated": false,
    "automatedBy": null
  },
  "extendedEntities": {
    "media": [
      {
        "additional_media_info": {
          "monetizable": false
        },
        "display_url": "pic.twitter.com/txQd2wdKz0",
        "expanded_url": "https://twitter.com/ManlingLi_/status/2067697152722415623/video/1",
        "ext_master_playlist_only": [],
        "ext_media_availability": {
          "status": "Available"
        },
        "ext_playlists": [],
        "id_str": "2067696956114407424",
        "indices": [
          273,
          296
        ],
        "media_key": "13_2067696956114407424",
        "media_results": {
          "id": "QXBpTWVkaWFSZXN1bHRzOgwABAoAARyx720sWxAAAAA=",
          "result": {
            "__typename": "ApiMedia",
            "id": "QXBpTWVkaWE6DAAECgABHLHvbSxbEAAAAA==",
            "media_key": "13_2067696956114407424"
          }
        },
        "media_url_https": "https://pbs.twimg.com/amplify_video_thumb/2067696956114407424/img/ntjDAV1n-gf8vkJ_.jpg",
        "original_info": {
          "focus_rects": [],
          "height": 1080,
          "width": 1920
        },
        "sizes": {
          "large": {
            "h": 1080,
            "w": 1920
          }
        },
        "type": "video",
        "url": "https://t.co/txQd2wdKz0",
        "video_info": {
          "aspect_ratio": [
            16,
            9
          ],
          "duration_millis": 47120,
          "variants": [
            {
              "content_type": "application/x-mpegURL",
              "url": "https://video.twimg.com/amplify_video/2067696956114407424/pl/stpIsLR2ANU4yYUX.m3u8?tag=28"
            },
            {
              "bitrate": 256000,
              "content_type": "video/mp4",
              "url": "https://video.twimg.com/amplify_video/2067696956114407424/vid/avc1/480x270/kEtCFWebgEJ8Fi_r.mp4?tag=28"
            },
            {
              "bitrate": 832000,
              "content_type": "video/mp4",
              "url": "https://video.twimg.com/amplify_video/2067696956114407424/vid/avc1/640x360/fuyoJaP1jdsfDUUF.mp4?tag=28"
            },
            {
              "bitrate": 2176000,
              "content_type": "video/mp4",
              "url": "https://video.twimg.com/amplify_video/2067696956114407424/vid/avc1/1280x720/7dvQBet-UuGB4rjZ.mp4?tag=28"
            },
            {
              "bitrate": 10368000,
              "content_type": "video/mp4",
              "url": "https://video.twimg.com/amplify_video/2067696956114407424/vid/avc1/1920x1080/qdXJ7UpSfpqPlUDJ.mp4?tag=28"
            }
          ]
        }
      }
    ]
  },
  "card": null,
  "place": {},
  "entities": {
    "hashtags": [],
    "symbols": [],
    "timestamps": [],
    "urls": [],
    "user_mentions": [
      {
        "id_str": "1580959771829706754",
        "indices": [
          857,
          867
        ],
        "name": "Kangrui Wang",
        "screen_name": "James_KKW"
      },
      {
        "id_str": "761527442",
        "indices": [
          888,
          898
        ],
        "name": "Linjie (Lindsey) Li",
        "screen_name": "LINJIEFUN"
      },
      {
        "id_str": "1718102739065229312",
        "indices": [
          899,
          914
        ],
        "name": "Zhengyuan Yang",
        "screen_name": "zhengyuan_yang"
      },
      {
        "id_str": "1636543817741647872",
        "indices": [
          915,
          928
        ],
        "name": "Shiqi Chen",
        "screen_name": "shiqi_chen17"
      },
      {
        "id_str": "1507697433593098244",
        "indices": [
          929,
          936
        ],
        "name": "Zihan \"Zenus\" Wang",
        "screen_name": "wzenus"
      },
      {
        "id_str": "130745589",
        "indices": [
          937,
          946
        ],
        "name": "Fei-Fei Li",
        "screen_name": "drfeifei"
      },
      {
        "id_str": "27546246",
        "indices": [
          947,
          959
        ],
        "name": "Jiajun Wu",
        "screen_name": "jiajunwu_cs"
      },
      {
        "id_str": "1059680847425527808",
        "indices": [
          1012,
          1026
        ],
        "name": "Stanford AI Lab",
        "screen_name": "StanfordAILab"
      },
      {
        "id_str": "2832190776",
        "indices": [
          1027,
          1039
        ],
        "name": "Stanford Vision and Learning Lab",
        "screen_name": "StanfordSVL"
      },
      {
        "id_str": "21457289",
        "indices": [
          1040,
          1053
        ],
        "name": "Microsoft Research",
        "screen_name": "MSFTResearch"
      }
    ]
  },
  "quoted_tweet": null,
  "retweeted_tweet": null,
  "isLimitedReply": false,
  "communityInfo": null,
  "article": null
}