Upload 27 files

roundabout_vau_category_qa_pairs_test.json

@@ -0,0 +1,590 @@
+[
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nAnalyze the entire traffic video and output exactly one label token A, B, C, or D for the most severe anomaly, with no extra text."
+      },
+      {
+        "from": "gpt",
+        "value": "D"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_142_26_2.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect all road users across the whole clip and output exactly one label A, B, C, or D for the most severe and relevant anomaly, nothing else."
+      },
+      {
+        "from": "gpt",
+        "value": "C"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_27_30_2.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nReview the full video for anomalous driving and return only a single token A, B, C, or D that best matches the most severe event."
+      },
+      {
+        "from": "gpt",
+        "value": "C"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Gray-96TH-10-31-25.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect all road users across the whole clip and output exactly one label A, B, C, or D for the most severe and relevant anomaly, nothing else."
+      },
+      {
+        "from": "gpt",
+        "value": "C"
+      }
+    ],
+    "video": "/data/Motorcycle-Rndbt-Rangeline-Elm-2020.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWatch the complete video, choose the single best label A, B, C, or D based on the most severe anomaly present, and output only that token."
+      },
+      {
+        "from": "gpt",
+        "value": "D"
+      }
+    ],
+    "video": "/data/Truck-RT-Rangeline-Elm-10-21-25.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nAnalyze the entire traffic video and output exactly one label token A, B, C, or D for the most severe anomaly, with no extra text."
+      },
+      {
+        "from": "gpt",
+        "value": "C"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_180_18_3.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWatch the complete video, choose the single best label A, B, C, or D based on the most severe anomaly present, and output only that token."
+      },
+      {
+        "from": "gpt",
+        "value": "B"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_296_21_3.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nReview the full video for anomalous driving and return only a single token A, B, C, or D that best matches the most severe event."
+      },
+      {
+        "from": "gpt",
+        "value": "C"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_36_18_3.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nAnalyze the entire traffic video and output exactly one label token A, B, C, or D for the most severe anomaly, with no extra text."
+      },
+      {
+        "from": "gpt",
+        "value": "C"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_43_18_3.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nAnalyze the entire traffic video and output exactly one label token A, B, C, or D for the most severe anomaly, with no extra text."
+      },
+      {
+        "from": "gpt",
+        "value": "C"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_168_18_3.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect all road users across the whole clip and output exactly one label A, B, C, or D for the most severe and relevant anomaly, nothing else."
+      },
+      {
+        "from": "gpt",
+        "value": "D"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_216_27_1.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWatch the complete video, choose the single best label A, B, C, or D based on the most severe anomaly present, and output only that token."
+      },
+      {
+        "from": "gpt",
+        "value": "B"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_217_21_3.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nAnalyze the entire traffic video and output exactly one label token A, B, C, or D for the most severe anomaly, with no extra text."
+      },
+      {
+        "from": "gpt",
+        "value": "C"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_221_19_4.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nAnalyze the entire traffic video and output exactly one label token A, B, C, or D for the most severe anomaly, with no extra text."
+      },
+      {
+        "from": "gpt",
+        "value": "B"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_249_23_4.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect all road users across the whole clip and output exactly one label A, B, C, or D for the most severe and relevant anomaly, nothing else."
+      },
+      {
+        "from": "gpt",
+        "value": "C"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_303_18_3.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nReview the full video for anomalous driving and return only a single token A, B, C, or D that best matches the most severe event."
+      },
+      {
+        "from": "gpt",
+        "value": "C"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_312_18_3.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect all road users across the whole clip and output exactly one label A, B, C, or D for the most severe and relevant anomaly, nothing else."
+      },
+      {
+        "from": "gpt",
+        "value": "B"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_364_21_3.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nReview the full video for anomalous driving and return only a single token A, B, C, or D that best matches the most severe event."
+      },
+      {
+        "from": "gpt",
+        "value": "C"
+      }
+    ],
+    "video": "/data/Car-Bike-Rangeline-CarmelDr-10-16-25.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nAnalyze the entire traffic video and output exactly one label token A, B, C, or D for the most severe anomaly, with no extra text."
+      },
+      {
+        "from": "gpt",
+        "value": "C"
+      }
+    ],
+    "video": "/data/Car-Car-L1-Towne-106TH-02-05-25.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWatch the complete video, choose the single best label A, B, C, or D based on the most severe anomaly present, and output only that token."
+      },
+      {
+        "from": "gpt",
+        "value": "C"
+      }
+    ],
+    "video": "/data/Car-Car-L2-Springmill-Main-08-09-24.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWatch the complete video, choose the single best label A, B, C, or D based on the most severe anomaly present, and output only that token."
+      },
+      {
+        "from": "gpt",
+        "value": "C"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Rangeline-CityCenter-11-01-24.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nDetermine if any anomalies occur in the video and output only one token A, B, C, or D corresponding to the most severe anomaly observed."
+      },
+      {
+        "from": "gpt",
+        "value": "C"
+      }
+    ],
+    "video": "/data/CarR3.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nDetermine if any anomalies occur in the video and output only one token A, B, C, or D corresponding to the most severe anomaly observed."
+      },
+      {
+        "from": "gpt",
+        "value": "C"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Cnstrct-Towne-Main-07-17-25.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nDetermine if any anomalies occur in the video and output only one token A, B, C, or D corresponding to the most severe anomaly observed."
+      },
+      {
+        "from": "gpt",
+        "value": "C"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Springmill-106TH-4-12-24.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nDetermine if any anomalies occur in the video and output only one token A, B, C, or D corresponding to the most severe anomaly observed."
+      },
+      {
+        "from": "gpt",
+        "value": "B"
+      }
+    ],
+    "video": "/data/CarStall.mov",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nReview the full video for anomalous driving and return only a single token A, B, C, or D that best matches the most severe event."
+      },
+      {
+        "from": "gpt",
+        "value": "D"
+      }
+    ],
+    "video": "/data/Trailer-Fall-HazelDell-116TH-12-04-23.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nAnalyze the entire traffic video and output exactly one label token A, B, C, or D for the most severe anomaly, with no extra text."
+      },
+      {
+        "from": "gpt",
+        "value": "D"
+      }
+    ],
+    "video": "/data/Truck-TR-Rangeline-Elm-2.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWatch the complete video, choose the single best label A, B, C, or D based on the most severe anomaly present, and output only that token."
+      },
+      {
+        "from": "gpt",
+        "value": "C"
+      }
+    ],
+    "video": "/data/Car-Car-PE-HazelDell-116TH-1-28-26.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nReview the full video for anomalous driving and return only a single token A, B, C, or D that best matches the most severe event."
+      },
+      {
+        "from": "gpt",
+        "value": "C"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Rangeline-136TH-1-2-26.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWatch the complete video, choose the single best label A, B, C, or D based on the most severe anomaly present, and output only that token."
+      },
+      {
+        "from": "gpt",
+        "value": "B"
+      }
+    ],
+    "video": "/data/no4_cat23_sev4.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nDetermine if any anomalies occur in the video and output only one token A, B, C, or D corresponding to the most severe anomaly observed."
+      },
+      {
+        "from": "gpt",
+        "value": "B"
+      }
+    ],
+    "video": "/data/no11_cat23_sev4.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWatch the complete video, choose the single best label A, B, C, or D based on the most severe anomaly present, and output only that token."
+      },
+      {
+        "from": "gpt",
+        "value": "B"
+      }
+    ],
+    "video": "/data/no6_cat23_sev4.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nAnalyze the entire traffic video and output exactly one label token A, B, C, or D for the most severe anomaly, with no extra text."
+      },
+      {
+        "from": "gpt",
+        "value": "B"
+      }
+    ],
+    "video": "/data/no22_cat22_sev4.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nAnalyze the entire traffic video and output exactly one label token A, B, C, or D for the most severe anomaly, with no extra text."
+      },
+      {
+        "from": "gpt",
+        "value": "B"
+      }
+    ],
+    "video": "/data/no21_cat23_sev4.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect all road users across the whole clip and output exactly one label A, B, C, or D for the most severe and relevant anomaly, nothing else."
+      },
+      {
+        "from": "gpt",
+        "value": "A"
+      }
+    ],
+    "video": "/data/N-HazelDell-116TH-C2-Night_537-1008.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nAnalyze the entire traffic video and output exactly one label token A, B, C, or D for the most severe anomaly, with no extra text."
+      },
+      {
+        "from": "gpt",
+        "value": "A"
+      }
+    ],
+    "video": "/data/N-HazelDell-Main-C1-Day-Cloudy_6856-7175.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWatch the complete video, choose the single best label A, B, C, or D based on the most severe anomaly present, and output only that token."
+      },
+      {
+        "from": "gpt",
+        "value": "A"
+      }
+    ],
+    "video": "/data/N-HazelDell-Main-C2-Night_6881-6981.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWatch the complete video, choose the single best label A, B, C, or D based on the most severe anomaly present, and output only that token."
+      },
+      {
+        "from": "gpt",
+        "value": "A"
+      }
+    ],
+    "video": "/data/N-Illinois-Main-C2-Day-Sunny_2628-3261.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nDetermine if any anomalies occur in the video and output only one token A, B, C, or D corresponding to the most severe anomaly observed."
+      },
+      {
+        "from": "gpt",
+        "value": "A"
+      }
+    ],
+    "video": "/data/N-Springmill-Main-C2-Day-Sunny_3533-3926.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWatch the complete video, choose the single best label A, B, C, or D based on the most severe anomaly present, and output only that token."
+      },
+      {
+        "from": "gpt",
+        "value": "A"
+      }
+    ],
+    "video": "/data/frames-000000_s3859_n421.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWatch the complete video, choose the single best label A, B, C, or D based on the most severe anomaly present, and output only that token."
+      },
+      {
+        "from": "gpt",
+        "value": "A"
+      }
+    ],
+    "video": "/data/frames-000042_s2243_n364.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWatch the complete video, choose the single best label A, B, C, or D based on the most severe anomaly present, and output only that token."
+      },
+      {
+        "from": "gpt",
+        "value": "A"
+      }
+    ],
+    "video": "/data/frames-000046_s4898_n538.mp4",
+    "system_prompt": "\n        You are a video classification system that identifies anomalous driving behavior in traffic videos.\n\n        You must assign exactly one label token from the set {A, B, C, D} based on all visible road users across the entire video. Choose the single best label using the definitions below. If multiple anomalies occur, output only the most severe and more relevant one.\n        \n        Labels (output exactly one token):\n\n        - A: No anomaly. All visible road users behave normally.\n        - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n        - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n        - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n        When you answer, you MUST:\n        1. Watch the entire video before deciding.\n        2. Output exactly one label token only: A or B or C or D.\n        3. Output nothing else: no words, no punctuation, no extra whitespace, no newlines.\n        "
+  }
+]

roundabout_vau_environment_qa_pairs_test.json

@@ -0,0 +1,590 @@
+[
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'night', 'weather': 'unknown', 'road_surface': 'unknown', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_142_26_2.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'night', 'weather': 'unknown', 'road_surface': 'unknown', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_27_30_2.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Gray-96TH-10-31-25.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'dawn_dusk', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'single_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/Motorcycle-Rndbt-Rangeline-Elm-2020.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'single_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/Truck-RT-Rangeline-Elm-10-21-25.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'clear', 'road_surface': 'unknown', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_180_18_3.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'clear', 'road_surface': 'unknown', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_296_21_3.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_36_18_3.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'night', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_43_18_3.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_168_18_3.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_216_27_1.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_217_21_3.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_221_19_4.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'night', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_249_23_4.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_303_18_3.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_312_18_3.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_364_21_3.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/Car-Bike-Rangeline-CarmelDr-10-16-25.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/Car-Car-L1-Towne-106TH-02-05-25.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/Car-Car-L2-Springmill-Main-08-09-24.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Rangeline-CityCenter-11-01-24.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'cloudy', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/CarR3.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'cloudy', 'road_surface': 'dry', 'road_type': 'multi_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Cnstrct-Towne-Main-07-17-25.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'night', 'weather': 'rainy', 'road_surface': 'wet', 'road_type': 'single_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Springmill-106TH-4-12-24.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'snowy', 'road_surface': 'wet', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/CarStall.mov",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/Trailer-Fall-HazelDell-116TH-12-04-23.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'single_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/Truck-TR-Rangeline-Elm-2.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'clear', 'road_surface': 'wet', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/Car-Car-PE-HazelDell-116TH-1-28-26.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'single_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Rangeline-136TH-1-2-26.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/no4_cat23_sev4.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'night', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/no11_cat23_sev4.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'night', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/no6_cat23_sev4.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'cloudy', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/no22_cat22_sev4.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'night', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/no21_cat23_sev4.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'night', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/N-HazelDell-116TH-C2-Night_537-1008.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'cloudy', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/N-HazelDell-Main-C1-Day-Cloudy_6856-7175.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'night', 'weather': 'cloudy', 'road_surface': 'unknown', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/N-HazelDell-Main-C2-Night_6881-6981.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/N-Illinois-Main-C2-Day-Sunny_2628-3261.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/N-Springmill-Main-C2-Day-Sunny_3533-3926.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'day', 'weather': 'clear', 'road_surface': 'dry', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/frames-000000_s3859_n421.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'night', 'weather': 'unknown', 'road_surface': 'unknown', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/frames-000042_s2243_n364.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nExtract the environment factors (time of day, weather, road surface, road type) from the video and provide them in a structured format."
+      },
+      {
+        "from": "gpt",
+        "value": "{'time_of_day': 'night', 'weather': 'unknown', 'road_surface': 'unknown', 'road_type': 'double_lane_roundabout'}"
+      }
+    ],
+    "video": "/data/frames-000046_s4898_n538.mp4",
+    "system_prompt": "\n            You will be given a short traffic video clip.\n\n            Task: Extract ONLY the environment_factor fields and output ONLY valid JSON (no extra text).\n\n            Use exactly these allowed values:\n            - time_of_day: [\"day\",\"night\",\"dawn_dusk\",\"unknown\"]\n            - weather: [\"clear\",\"cloudy\",\"rainy\",\"snowy\",\"foggy\",\"unknown\"]\n            - road_surface: [\"dry\",\"wet\",\"snow_covered\",\"unknown\"]\n            - road_type: [\"single_lane_roundabout\",\"double_lane_roundabout\",\"multi_lane_roundabout\",\"unknown\"]\n\n            Rules:\n            - Use only visible evidence. If not observable, use \"unknown\".\n            - If the scene is not clearly a roundabout, set road_type to \"unknown\".\n\n            Output JSON schema (exact keys):\n            {\n            \"environment_factor\": {\n                \"time_of_day\": \"...\",\n                \"weather\": \"...\",\n                \"road_surface\": \"...\",\n                \"road_type\": \"...\"\n            }\n            }\n        "
+  }
+]

roundabout_vau_simple_description_qa_pairs_test.json

@@ -0,0 +1,444 @@
+[
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you describe the anomaly events observed in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A white sedan stops in the outer circulating lane of the roundabout to yield to a pedestrian crossing through the roadway."
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_142_26_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you describe the anomaly events observed in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A silver-gray SUV entering the roundabout brakes hard in the inner entry lane as it approaches circulating traffic."
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_27_30_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nHow would you detail the anomaly events found in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A light-colored sedan enters the roundabout without yielding and collides with a black sedan already circulating, pushing it toward the central island."
+      }
+    ],
+    "video": "/data/Car-Car-PE-Gray-96TH-10-31-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you specify the anomaly events present in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A motorcycle loses control in the roundabout and crashes onto the central island."
+      }
+    ],
+    "video": "/data/Motorcycle-Rndbt-Rangeline-Elm-2020.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nProvide a summary of the anomaly events in the video."
+      },
+      {
+        "from": "gpt",
+        "value": "A white truck mounts the curb/sidewalk while exiting the roundabout and strikes a lamp post."
+      }
+    ],
+    "video": "/data/Truck-RT-Rangeline-Elm-10-21-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nHow would you describe the particular anomaly events in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A dark-colored SUV enters into the circulating lane and cuts off a black SUV already in the roundabout, forcing it to slow to avoid a conflict."
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_180_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nGive a description of the detected anomaly events in this video."
+      },
+      {
+        "from": "gpt",
+        "value": "A dark SUV makes an improper diagonal turn within the roundabout to reach an exit, not following the lane markings."
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_296_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nHow would you detail the anomaly events found in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A light-colored sedan enters the roundabout without yielding and cuts off a black SUV already circulating, forcing the SUV to slow down to avoid a collision."
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_36_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWhat details can you provide about the anomaly in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A dark sedan enters the roundabout without yielding and cuts in front of a white SUV, forcing the SUV to slow and briefly stop to avoid a collision."
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_43_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWhat details can you provide about the anomaly in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A black sedan merges into the roundabout without yielding, forcing two red SUVs already circulating to slow down to avoid a collision."
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_168_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nProvide a summary of the anomaly events in the video."
+      },
+      {
+        "from": "gpt",
+        "value": "A dark sedan stops at the roundabout entry/crosswalk area to yield to a pedestrian, then continues forward."
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_216_27_1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you provide a summary of the anomaly events in this video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A dark sedan enters from the inner entry lane and makes an illegal early right exit from the roundabout."
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_217_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nHow would you describe the particular anomaly events in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A light-colored sedan merges into the inner circulating lane with insufficient gap, forcing a black SUV to slow down to avoid a collision."
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_221_19_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you give a description of the anomaly events in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A white sedan enters the roundabout through an exit lane (wrong way) and then continues normally."
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_249_23_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you give a description of the anomaly events in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A white pickup merges into the roundabout and cuts in front of a circulating white SUV, forcing the SUV to slow down to avoid a collision."
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_303_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nGive a description of the detected anomaly events in this video."
+      },
+      {
+        "from": "gpt",
+        "value": "A white pickup slows and briefly stops in the inner circulating lane, forcing a following white car to brake and stop inside the roundabout."
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_312_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nHow would you detail the anomaly events found in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A dark sedan changes lanes inward in the roundabout instead of exiting from the outer lane."
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_364_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nGive a description of the detected anomaly events in this video."
+      },
+      {
+        "from": "gpt",
+        "value": "A dark sedan makes slight contact with a cyclist in the crosswalk, causing the cyclist to stumble."
+      }
+    ],
+    "video": "/data/Car-Bike-Rangeline-CarmelDr-10-16-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nProvide a summary of the anomaly events in the video."
+      },
+      {
+        "from": "gpt",
+        "value": "A black sedan cuts across lanes near the first exit and collides with a red sedan in the outer circulating lane."
+      }
+    ],
+    "video": "/data/Car-Car-L1-Towne-106TH-02-05-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you specify the anomaly events present in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A white SUV changes lanes toward the exit and collides with a black sedan near the right exit of the roundabout."
+      }
+    ],
+    "video": "/data/Car-Car-L2-Springmill-Main-08-09-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nHow would you describe the particular anomaly events in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A black SUV merges into the roundabout without yielding, causing a black sedan to brake and a following white sedan to rear-end it."
+      }
+    ],
+    "video": "/data/Car-Car-PE-Rangeline-CityCenter-11-01-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you provide a summary of the anomaly events in this video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A silver SUV makes an illegal left turn across the exit area and collides with a red sedan that is exiting."
+      }
+    ],
+    "video": "/data/CarR3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nHow would you detail the anomaly events found in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A black sedan enters the roundabout and drives into a construction hole at the outer entry lane, coming to a stop."
+      }
+    ],
+    "video": "/data/Car-Rndbt-Cnstrct-Towne-Main-07-17-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you specify the anomaly events present in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A sedan drives into the central island of the roundabout and collides with the decorations before reversing back and exiting."
+      }
+    ],
+    "video": "/data/Car-Rndbt-Springmill-106TH-4-12-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you describe the anomaly events observed in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A black SUV pauses in the exit lane and reverses back into the roundabout before continuing forward."
+      }
+    ],
+    "video": "/data/CarStall.mov"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWhat details can you provide about the anomaly in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A silver-gray pickup towing a black trailer overturns in the roundabout and comes to rest in the circulating lane."
+      }
+    ],
+    "video": "/data/Trailer-Fall-HazelDell-116TH-12-04-23.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nHow would you describe the particular anomaly events in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A red HGV drives onto the roundabout's central island grass area."
+      }
+    ],
+    "video": "/data/Truck-TR-Rangeline-Elm-2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nHow would you detail the anomaly events found in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A blue sedan enters the roundabout without yielding and rear-ends a black SUV that is already circulating."
+      }
+    ],
+    "video": "/data/Car-Car-PE-HazelDell-116TH-1-28-26.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nHow would you describe the particular anomaly events in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A dark sedan leaves the roadway near the top-left exit and collides with a yield sign post before stopping in the entry lane."
+      }
+    ],
+    "video": "/data/Car-Rndbt-Rangeline-136TH-1-2-26.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nHow would you describe the particular anomaly events in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A car drives toward the roundabout from the inner exit lane against the normal traffic direction."
+      }
+    ],
+    "video": "/data/no4_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you give a description of the anomaly events in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A black SUV enters from an exit lane and drives counterclockwise against the roundabout flow before exiting to the right."
+      }
+    ],
+    "video": "/data/no11_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you describe the anomaly events observed in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A silver car drives the wrong way in the outer lane and exits the roundabout via the entry lane."
+      }
+    ],
+    "video": "/data/no6_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWhat details can you provide about the anomaly in the video?"
+      },
+      {
+        "from": "gpt",
+        "value": "A silver sedan drives straight across the roundabout's central island against the circulating direction and exits to the right."
+      }
+    ],
+    "video": "/data/no22_cat22_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nProvide a summary of the anomaly events in the video."
+      },
+      {
+        "from": "gpt",
+        "value": "A silver sedan enters the roundabout and immediately drives clockwise (wrong-way) instead of circulating counterclockwise."
+      }
+    ],
+    "video": "/data/no21_cat23_sev4.mp4"
+  }
+]

roundabout_vau_simple_reasoning_qa_pairs_test.json

@@ -0,0 +1,444 @@
+[
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWhat evidence do you have to support your judgment of this event as an anomaly?"
+      },
+      {
+        "from": "gpt",
+        "value": "The sedan stops in an active circulating lane of the roundabout, creating an abnormal obstruction to traffic flow while yielding to a pedestrian."
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_142_26_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you provide the reasons for considering this event as abnormal?"
+      },
+      {
+        "from": "gpt",
+        "value": "The SUV brakes hard while approaching the roundabout with other vehicles circulating, creating a near-collision risk."
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_27_30_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWhat evidence do you have to support your judgment of this event as an anomaly?"
+      },
+      {
+        "from": "gpt",
+        "value": "The entering sedan fails to yield to a circulating sedan, causing a collision in the roundabout."
+      }
+    ],
+    "video": "/data/Car-Car-PE-Gray-96TH-10-31-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you share your analysis of the anomalous event?"
+      },
+      {
+        "from": "gpt",
+        "value": "The motorcycle loses control and crashes, resulting in the rider being thrown onto the central island."
+      }
+    ],
+    "video": "/data/Motorcycle-Rndbt-Rangeline-Elm-2020.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWhat evidence do you have to support your judgment of this event as an anomaly?"
+      },
+      {
+        "from": "gpt",
+        "value": "The truck drives onto the curb/sidewalk while exiting and hits roadside infrastructure."
+      }
+    ],
+    "video": "/data/Truck-RT-Rangeline-Elm-10-21-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you share your analysis of the anomalous event?"
+      },
+      {
+        "from": "gpt",
+        "value": "The entering SUV fails to yield and cuts into the circulating lane, forcing the circulating SUV to slow to avoid a near-collision."
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_180_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCan you provide the reasons for considering it anomalous?"
+      },
+      {
+        "from": "gpt",
+        "value": "The SUV violates lane discipline by cutting diagonally within the roundabout to take an exit instead of exiting properly from its lane."
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_296_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCan you analyze the factors contributing to this anomalous event?"
+      },
+      {
+        "from": "gpt",
+        "value": "The entering sedan fails to yield to the circulating SUV and cuts in front, forcing the SUV to slow down to avoid a collision."
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_36_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you provide the reasons for considering this event as abnormal?"
+      },
+      {
+        "from": "gpt",
+        "value": "The sedan fails to yield when entering the roundabout and cuts off the circulating SUV, causing the SUV to slow and briefly stop to avoid a collision."
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_43_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCan you provide the reasons for considering it anomalous?"
+      },
+      {
+        "from": "gpt",
+        "value": "The black sedan enters the circulating lane without yielding to vehicles already in the roundabout, causing them to slow to avoid a collision."
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_168_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you provide the reasons for considering this event as abnormal?"
+      },
+      {
+        "from": "gpt",
+        "value": "The sedan stops in the active entry lane at the crosswalk area to yield to a pedestrian, temporarily blocking the lane."
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_216_27_1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWhat evidence do you have to support your judgment of this event as an anomaly?"
+      },
+      {
+        "from": "gpt",
+        "value": "The sedan exits to the right from the inner entry/inner circulating path instead of continuing to circulate to a proper exit."
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_217_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you share your analysis of the anomalous event?"
+      },
+      {
+        "from": "gpt",
+        "value": "The sedan makes an unsafe lane change into the inner circulating lane, forcing the SUV already there to slow down to avoid a near collision."
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_221_19_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you share your analysis of the anomalous event?"
+      },
+      {
+        "from": "gpt",
+        "value": "The sedan enters the roundabout via an exit lane instead of the designated entry lane, violating the correct direction of travel."
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_249_23_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCan you analyze the factors contributing to this anomalous event?"
+      },
+      {
+        "from": "gpt",
+        "value": "The pickup merges into the circulating lane without sufficient gap and cuts off the white SUV, causing it to slow down to avoid a near collision."
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_303_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCan you analyze the factors contributing to this anomalous event?"
+      },
+      {
+        "from": "gpt",
+        "value": "The pickup stops in the circulating lane and causes a following vehicle to brake and stop to avoid a collision."
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_312_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWhat evidence do you have to support your judgment of this event as an anomaly?"
+      },
+      {
+        "from": "gpt",
+        "value": "The sedan crosses the lane markings from the outer circulating lane into the inner circulating lane to continue circulating instead of exiting."
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_364_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you provide the reasons for considering this event as abnormal?"
+      },
+      {
+        "from": "gpt",
+        "value": "The sedan makes contact with a cyclist in the crosswalk, causing the cyclist to stumble."
+      }
+    ],
+    "video": "/data/Car-Bike-Rangeline-CarmelDr-10-16-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWhat evidence do you have to support your judgment of this event as an anomaly?"
+      },
+      {
+        "from": "gpt",
+        "value": "The black sedan makes an unsafe right turn across lanes into the outer circulating lane, causing a collision with the red sedan."
+      }
+    ],
+    "video": "/data/Car-Car-L1-Towne-106TH-02-05-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCan you provide the reasons for considering it anomalous?"
+      },
+      {
+        "from": "gpt",
+        "value": "The SUV makes an unsafe lane change toward the exit and hits the sedan."
+      }
+    ],
+    "video": "/data/Car-Car-L2-Springmill-Main-08-09-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCan you provide the reasons for considering it anomalous?"
+      },
+      {
+        "from": "gpt",
+        "value": "The black SUV fails to yield when entering the roundabout, leading to a rear-end collision after the circulating black sedan brakes."
+      }
+    ],
+    "video": "/data/Car-Car-PE-Rangeline-CityCenter-11-01-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWhat evidence do you have to support your judgment of this event as an anomaly?"
+      },
+      {
+        "from": "gpt",
+        "value": "The SUV cuts across the exit lane while turning left inside the roundabout, causing a collision with the exiting sedan."
+      }
+    ],
+    "video": "/data/CarR3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWhat evidence do you have to support your judgment of this event as an anomaly?"
+      },
+      {
+        "from": "gpt",
+        "value": "The sedan drives into a construction hole at the roundabout entry and becomes stopped due to the impact."
+      }
+    ],
+    "video": "/data/Car-Rndbt-Cnstrct-Towne-Main-07-17-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you share your analysis of the anomalous event?"
+      },
+      {
+        "from": "gpt",
+        "value": "The sedan fails to follow the circulating lane and collides with objects in the central island."
+      }
+    ],
+    "video": "/data/Car-Rndbt-Springmill-106TH-4-12-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCan you analyze the factors contributing to this anomalous event?"
+      },
+      {
+        "from": "gpt",
+        "value": "The SUV reverses in the exit lane/roundabout, which is an illegal and unsafe maneuver."
+      }
+    ],
+    "video": "/data/CarStall.mov"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCan you analyze the factors contributing to this anomalous event?"
+      },
+      {
+        "from": "gpt",
+        "value": "The pickup overturns and stops in an active circulating lane of the roundabout."
+      }
+    ],
+    "video": "/data/Trailer-Fall-HazelDell-116TH-12-04-23.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you provide the reasons for considering this event as abnormal?"
+      },
+      {
+        "from": "gpt",
+        "value": "The HGV leaves the paved circulating lane and drives onto the central island grass area."
+      }
+    ],
+    "video": "/data/Truck-TR-Rangeline-Elm-2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWhat evidence do you have to support your judgment of this event as an anomaly?"
+      },
+      {
+        "from": "gpt",
+        "value": "The blue sedan fails to yield when entering the roundabout and collides with the rear of the circulating black SUV."
+      }
+    ],
+    "video": "/data/Car-Car-PE-HazelDell-116TH-1-28-26.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCan you analyze the factors contributing to this anomalous event?"
+      },
+      {
+        "from": "gpt",
+        "value": "The sedan departs the roadway and strikes a yield sign post near the exit."
+      }
+    ],
+    "video": "/data/Car-Rndbt-Rangeline-136TH-1-2-26.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCan you analyze the factors contributing to this anomalous event?"
+      },
+      {
+        "from": "gpt",
+        "value": "A car travels toward the roundabout from an exit lane in the wrong direction."
+      }
+    ],
+    "video": "/data/no4_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWhat evidence do you have to support your judgment of this event as an anomaly?"
+      },
+      {
+        "from": "gpt",
+        "value": "The SUV enters the roundabout via an exit lane and travels counterclockwise against the intended direction of circulation."
+      }
+    ],
+    "video": "/data/no11_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you provide the reasons for considering this event as abnormal?"
+      },
+      {
+        "from": "gpt",
+        "value": "The car travels the wrong way in the outer circulating lane and exits using the entry lane, violating the required direction of travel."
+      }
+    ],
+    "video": "/data/no6_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nWhat evidence do you have to support your judgment of this event as an anomaly?"
+      },
+      {
+        "from": "gpt",
+        "value": "The sedan drives across the central island instead of following the roundabout circulation direction."
+      }
+    ],
+    "video": "/data/no22_cat22_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you share your analysis of the anomalous event?"
+      },
+      {
+        "from": "gpt",
+        "value": "The sedan circulates clockwise in the roundabout, which is a wrong-way direction violation."
+      }
+    ],
+    "video": "/data/no21_cat23_sev4.mp4"
+  }
+]

roundabout_vau_summary_qa_pairs_test.json

@@ -0,0 +1,478 @@
+[
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nProvide a summary of the anomaly events in the video.\nPrior Category: D"
+      },
+      {
+        "from": "gpt",
+        "value": "At night at a lit two-lane roundabout, a white sedan travels from the upper-center entry into the outer circulating lane and then stops in the roundabout to wait for a pedestrian coming from the center-left sidewalk to cross through the outer circulating lane toward the crosswalk area. This mid-roundabout stop to yield to a person creates an unusual, potentially confusing obstruction for following vehicles in the live traffic stream."
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_142_26_2.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you give a summary of the anomaly events in the video?\nPrior Category: C"
+      },
+      {
+        "from": "gpt",
+        "value": "At night at a well-lit double-lane roundabout, the subject is a silver-gray SUV entering from the lower left approach in the inner entry lane and moving toward the circular junction. As it nears the roundabout with other vehicles circulating, the SUV brakes hard in the inner entry lane, creating a near-collision risk, then continues and exits toward the upper left."
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_27_30_2.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nProvide a summary of the anomaly events in the video.\nPrior Category: C"
+      },
+      {
+        "from": "gpt",
+        "value": "In daylight at a double-lane roundabout, a light-colored sedan approaches from the top-left and merges into the roundabout from the inner entry lane. It fails to yield to a black sedan already circulating and collides with it in the outer circulating lane. The black sedan is pushed toward the central island, and both vehicles then come to rest inside the roundabout."
+      }
+    ],
+    "video": "/data/Car-Car-PE-Gray-96TH-10-31-25.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nGive a summary of the detected anomaly events in this video.\nPrior Category: C"
+      },
+      {
+        "from": "gpt",
+        "value": "In the evening under dry conditions, a motorcycle enters a single-lane roundabout from the bottom-left approach. The rider loses control and is involved in a head-on collision, after which the motorcycle comes to a stop on the central island. The rider is then seen lying in the central grass area of the roundabout. This is a collision event caused by the motorcycle losing control."
+      }
+    ],
+    "video": "/data/Motorcycle-Rndbt-Rangeline-Elm-2020.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nSummarize the anomaly events observed in the video.\nPrior Category: D"
+      },
+      {
+        "from": "gpt",
+        "value": "In daylight under dry conditions, a white truck enters a single-lane roundabout from the bottom-right approach. As it exits via the mid-right first exit, part of the truck rides up onto the curb/sidewalk and the vehicle body strikes a lamp post, knocking it down. The truck then continues and leaves the scene via the right exit. This is abnormal road use because the vehicle mounts the curb/sidewalk and hits roadside infrastructure."
+      }
+    ],
+    "video": "/data/Truck-RT-Rangeline-Elm-10-21-25.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you give a summary of the anomaly events in the video?\nPrior Category: C"
+      },
+      {
+        "from": "gpt",
+        "value": "In daylight at a busy double-lane roundabout, a dark-colored SUV appears from the lower left at the inner entry lane and proceeds into the inner circulating lane. As it enters, it cuts off a black SUV already circulating in the roundabout, forcing that SUV to slow and adjust to avoid a conflict."
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_180_18_3.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nProvide a summary of the anomaly events in the video.\nPrior Category: B"
+      },
+      {
+        "from": "gpt",
+        "value": "In daylight at a two-lane roundabout, a dark SUV near the top center of the frame enters from the outer entry lane and travels into the circulating lane. Instead of exiting in a standard way, the SUV performs an improper diagonal turn within the roundabout toward an inner exit lane, not following the lane markings. This illegal maneuver briefly places it at an unusual angle relative to the lanes and could create confusion or conflict with nearby traffic."
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_296_21_3.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you give a summary of the anomaly events in the video?\nPrior Category: C"
+      },
+      {
+        "from": "gpt",
+        "value": "At a busy double-lane roundabout in daylight, the subject vehicle is a light-colored sedan that travels from the lower-center of the frame toward the roundabout via the outer entry lane. When it merges into the outer circulating lane, it fails to yield to a black SUV already circulating and cuts across its path, forcing the SUV to slow down to avoid a potential collision."
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_36_18_3.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you summarize the anomaly events observed in the video?\nPrior Category: C"
+      },
+      {
+        "from": "gpt",
+        "value": "At night at a double-lane roundabout, the subject is a dark sedan entering from the lower left approach in the inner entry lane and moving toward the circulating lane. As it nears the yield markings, a white SUV is approaching from the upper left enter lane toward the same area. Instead of waiting, the dark sedan continues into the roundabout and merges into the circulating lane directly in front of the white SUV, forcing the SUV to slow and briefly stop to avoid a collision, creating an unsafe conflict."
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_43_18_3.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you give a summary of the anomaly events in the video?\nPrior Category: C"
+      },
+      {
+        "from": "gpt",
+        "value": "On a sunny day at a busy two-lane roundabout, the black sedan starts from the mid-top of the frame and waits to merge into the roundabout in the yielding lane. When it merges into the inner circulating lane, it fails to yield to two red SUVs already circulating in the roundabout and forces them to slow down to avoid a potential collision."
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_168_18_3.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you summarize the anomaly events observed in the video?\nPrior Category: D"
+      },
+      {
+        "from": "gpt",
+        "value": "At a busy double-lane roundabout in clear daytime conditions on a dry road, a dark sedan approaches from the lower-left toward the entry crosswalk in the inner entry lane. As a pedestrian walks across from right to left in the marked crosswalk, the driver slows and stops at the crosswalk area, waiting while the person completes the crossing. Once the pedestrian is clear, the car continues forward toward the roundabout, with the anomaly being the vehicle stopping in an active entry lane and briefly blocking traffic flow while yielding."
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_216_27_1.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nSummarize the anomaly events observed in the video.\nPrior Category: B"
+      },
+      {
+        "from": "gpt",
+        "value": "On a sunny day at a busy two-lane roundabout, a dark sedan enters from the lower-left inner entry lane and moves into the inner circulating lane. Instead of continuing to circulate, the car makes an illegal early right turn and exits toward the mid-right side of the scene, violating proper roundabout lane/exit usage."
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_217_21_3.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nGive a summary of the detected anomaly events in this video.\nPrior Category: C"
+      },
+      {
+        "from": "gpt",
+        "value": "In daylight at a two-lane roundabout, a light-coloured sedan from the outer entry lane (center of frame) drives into the roundabout and then changes into the inner circulating lane. A black SUV is already approaching from the inner entry lane and continues in the inner circulating lane, but the sedan still merges with insufficient gap, forcing the SUV to slow down and leaving a very small gap between the two vehicles, creating a near-collision risk."
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_221_19_4.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nProvide a summary of the anomaly events in the video.\nPrior Category: B"
+      },
+      {
+        "from": "gpt",
+        "value": "At night at a two-lane roundabout, a white sedan first appears near the upper center of the frame and enters the roundabout through the exit lane rather than the entry lane (wrong way). After that, the vehicle continues normally and heads toward the inner exit lane on the center-right. This is a direction/maneuver violation because it enters against the intended roundabout flow."
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_249_23_4.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nProvide a summary of the anomaly events in the video.\nPrior Category: C"
+      },
+      {
+        "from": "gpt",
+        "value": "In daytime at a busy double-lane roundabout, the subject vehicle is the white pickup that starts at the lower left entry and waits in the yielding lane to merge into the roundabout. As it enters, a dark sedan is already traveling in the roundabout alongside it. When the pickup merges into the inner circulating lane, it cuts in front of a white SUV already circulating, forcing the SUV to slow down to avoid a potential collision."
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_303_18_3.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nProvide a summary of the anomaly events in the video.\nPrior Category: C"
+      },
+      {
+        "from": "gpt",
+        "value": "On a sunny day at a busy two-lane roundabout, a white pickup on the top-right of the frame is already in the roundabout in the inner circulating lane and plans to exit. The white pickup slows down and briefly stops inside the roundabout, which forces another white car behind it to brake and stop inside the roundabout, creating a near-collision risk due to the sudden blockage in the circulating lane."
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_312_18_3.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nGive a summary of the detected anomaly events in this video.\nPrior Category: B"
+      },
+      {
+        "from": "gpt",
+        "value": "In daylight at a busy two-lane roundabout, a dark sedan enters from the upper-left into the outer entry lane and proceeds through the outer circulating lane. Instead of following the lane marking and exiting from the right exit, it makes an illegal lane change inward to keep circulating inside the roundabout (around 13–15s). Finally, it exits through the top exit."
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_364_21_3.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nGive a summary of the detected anomaly events in this video.\nPrior Category: C"
+      },
+      {
+        "from": "gpt",
+        "value": "Daylight and sunny, at a double-lane roundabout with a crosswalk. A dark sedan travels from the top-right toward the top-left, entering via the inner entry lane. A cyclist moves from the mid-top toward the top-left, crossing at the crosswalk. As the cyclist crosses, the dark sedan makes slight contact with the cyclist, causing the cyclist to stumble / partially fall but remain upright (still standing) afterward. The dark sedan then continues and exits the scene via the exit lane. A few seconds later, a person wearing a blue shirt approaches the cyclist. This is a near-collision/collision event because the sedan contacts the cyclist in the crosswalk."
+      }
+    ],
+    "video": "/data/Car-Bike-Rangeline-CarmelDr-10-16-25.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you give a summary of the anomaly events in the video?\nPrior Category: C"
+      },
+      {
+        "from": "gpt",
+        "value": "In daylight at a double-lane roundabout, a black sedan approaches from the top-right using the inner entry lane and merges into the roundabout. Near the first exit, it makes an illegal right turn across lanes and collides with a red sedan circulating normally in the outer lane. After the collision, both vehicles stop inside the roundabout, as the black sedan cuts across the outer circulating lane into the red sedan's path."
+      }
+    ],
+    "video": "/data/Car-Car-L1-Towne-106TH-02-05-25.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nSummarize the anomaly events observed in the video.\nPrior Category: C"
+      },
+      {
+        "from": "gpt",
+        "value": "In sunny daylight at a double-lane roundabout, a white SUV travels in the inner lane from the top-left toward the right. As it approaches the right exit, the SUV makes an illegal lane change toward the exit and collides with a black sedan near the exit lane. After the collision, both vehicles leave the scene."
+      }
+    ],
+    "video": "/data/Car-Car-L2-Springmill-Main-08-09-24.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nSummarize the anomaly events observed in the video.\nPrior Category: C"
+      },
+      {
+        "from": "gpt",
+        "value": "In sunny daylight at a double-lane roundabout, a black SUV enters from the lower-right and merges without yielding to a black sedan already circulating in the outer lane. The black sedan brakes abruptly to avoid the SUV. A white sedan coming from the mid-center inner entry lane does not anticipate the sudden braking, continues forward, and rear-ends the black sedan. The black sedan and white sedan stop inside the roundabout, while the black SUV continues toward the upper-right. The collision occurs because the entering SUV does not yield and the following white sedan fails to stop in time."
+      }
+    ],
+    "video": "/data/Car-Car-PE-Rangeline-CityCenter-11-01-24.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you summarize the anomaly events observed in the video?\nPrior Category: C"
+      },
+      {
+        "from": "gpt",
+        "value": "In daylight at a double-lane roundabout, a silver SUV enters from the top-left and circulates in the outer lane. As it approaches the mid-right exit, instead of exiting, it makes an illegal left turn to continue circulating, cutting across the exit area. A red sedan that is preparing to exit collides with the side of the silver SUV. Both vehicles then stop inside the roundabout."
+      }
+    ],
+    "video": "/data/CarR3.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you summarize the anomaly events observed in the video?\nPrior Category: C"
+      },
+      {
+        "from": "gpt",
+        "value": "In daylight, at a multi-lane roundabout that was under construction, a black sedan entered from the left outer entry lane. The vehicle drove into a construction hole at the entry to the roundabout and came to a stop there. There was no secondary impact, traffic disruption was low, and no emergency response was observed; the crash occurred because the car entered the construction hole."
+      }
+    ],
+    "video": "/data/Car-Rndbt-Cnstrct-Towne-Main-07-17-25.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nSummarize the anomaly events observed in the video.\nPrior Category: C"
+      },
+      {
+        "from": "gpt",
+        "value": "At night under wet condition, on a single-lane roundabout, a sedan entered from the bottom-left of the scene and drives toward the top-right. The vehicle travels very fast and directly drives into the central island of the roundabout. It collides with decorations in the central island. Later than, it reserved back to the road, drives following the roundabout and exit through the top-mid of the scene."
+      }
+    ],
+    "video": "/data/Car-Rndbt-Springmill-106TH-4-12-24.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nSummarize the anomaly events observed in the video.\nPrior Category: B"
+      },
+      {
+        "from": "gpt",
+        "value": "In daylight on a snowy double-lane roundabout, a black SUV enters from the top-middle approach and proceeds through the first exit at the top-left. Instead of exiting the scene, it pauses/stalls briefly in the exit lane, then reverses slightly back into the roundabout before moving forward again and continuing to drive. Reversing in a roundabout/exit lane is an illegal and unsafe maneuver."
+      }
+    ],
+    "video": "/data/CarStall.mov",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nGive a summary of the detected anomaly events in this video.\nPrior Category: D"
+      },
+      {
+        "from": "gpt",
+        "value": "In daylight and under dry conditions, a silver-gray pickup towing a black trailer entered a double-lane roundabout from the top-left approach via the inner entry lane. The vehicle lost control and overturned in the upper-center inner circulating lane, then came to rest in the upper-right circulating lane. There was no secondary impact, but the overturned vehicle remained stopped in an active lane, creating an abnormal road obstruction."
+      }
+    ],
+    "video": "/data/Trailer-Fall-HazelDell-116TH-12-04-23.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nSummarize the anomaly events observed in the video.\nPrior Category: D"
+      },
+      {
+        "from": "gpt",
+        "value": "In daylight and under dry conditions, a red heavy goods vehicle (HGV) entered the single-lane roundabout from the top mid approach. While navigating the roundabout, it left the roadway and drove onto the grassed central area. The vehicle then exited the scene via the mid-top exit. There was no secondary impact, traffic disruption was low, and no emergency response was observed."
+      }
+    ],
+    "video": "/data/Truck-TR-Rangeline-Elm-2.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nSummarize the anomaly events observed in the video.\nPrior Category: C"
+      },
+      {
+        "from": "gpt",
+        "value": "In sunny daylight at a double-lane roundabout with heavy snow visible, a blue sedan approaches from the upper-center yielding lane and enters the roundabout without yielding. It strikes the rear of a black SUV already circulating, causing the SUV to spin out, and the vehicles come to rest near the exit lane. The collision is caused by the blue sedan failing to yield to circulating traffic."
+      }
+    ],
+    "video": "/data/Car-Car-PE-HazelDell-116TH-1-28-26.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nProvide a summary of the anomaly events in the video.\nPrior Category: C"
+      },
+      {
+        "from": "gpt",
+        "value": "In daylight at a single-lane roundabout, a dark sedan enters from the mid-top approach and heads toward the first exit at the top-left. Near the top-left exit, it leaves the roadway, mounts the curb, and collides head-on with a yield sign post. The sedan then comes to rest in the top-left entry lane."
+      }
+    ],
+    "video": "/data/Car-Rndbt-Rangeline-136TH-1-2-26.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nSummarize the anomaly events observed in the video.\nPrior Category: B"
+      },
+      {
+        "from": "gpt",
+        "value": "At a sunny double-lane roundabout in a commercial area, a car travels from the upper-center inner exit lane toward the roundabout against the marked direction. This is a direction violation because the vehicle is driving the wrong way on the exit lane."
+      }
+    ],
+    "video": "/data/no4_cat23_sev4.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nProvide a summary of the anomaly events in the video.\nPrior Category: B"
+      },
+      {
+        "from": "gpt",
+        "value": "At night at a double-lane roundabout, a black SUV enters via an exit lane from the bottom-left, driving counterclockwise against the intended flow within the circulatory roadway, then leaves correctly through the exit lane on the right. This is a direction/maneuver violation because it enters from an exit and proceeds the wrong way around the roundabout."
+      }
+    ],
+    "video": "/data/no11_cat23_sev4.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nProvide a summary of the anomaly events in the video.\nPrior Category: B"
+      },
+      {
+        "from": "gpt",
+        "value": "At night at a double-lane roundabout, a silver car drives the wrong way on the outer lane and leaves the roundabout via the entry lane on the right side of the frame, which is a direction/maneuver violation."
+      }
+    ],
+    "video": "/data/no6_cat23_sev4.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nCould you summarize the anomaly events observed in the video?\nPrior Category: B"
+      },
+      {
+        "from": "gpt",
+        "value": "In a double-lane roundabout on a cloudy day with a dry road surface, a silver sedan starts from the central island in the middle of the frame, drives straight across the central island against the traffic flow, and then exits through the exit lane on the right. Other vehicles wait while it clears, as this is a direction violation."
+      }
+    ],
+    "video": "/data/no22_cat22_sev4.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nSummarize the anomaly events observed in the video.\nPrior Category: B"
+      },
+      {
+        "from": "gpt",
+        "value": "At a snowy double-lane roundabout at night, a silver sedan on the mid-bottom approaches the roundabout from the inner entry lane; after it enters, the vehicle directly turns left (clockwise) and continues wrong-way circulating toward the left side in the outer circulating lane until the end of the clip, violating the required counterclockwise roundabout direction."
+      }
+    ],
+    "video": "/data/no21_cat23_sev4.mp4",
+    "system_prompt": "\n            You are an expert driving behavior analyst for traffic surveillance videos.\n\n            Inputs you will receive:\n\n            - A short traffic video clip.\n\n            - A category label (A/B/C/D) describing the anomaly type.\n\n\n            Category meanings:\n\n            - A: No anomaly. All visible road users behave normally.\n\n            - B: Direction or maneuver violation, including wrong-way driving, illegal or unsafe turns, and illegal or unsafe lane changes.\n\n            - C: Near-collision or collision event, including cut-offs that force hard braking or evasive action, near-misses, and crashes.\n\n            - D: Abnormal road use, including stopping in an active lane or mid-road (breakdown, yielding to emergency vehicles, letting pedestrians cross) and driving or parking in non-vehicle areas (sidewalk, shoulder, etc.).\n\n\n            Task:\n\n            Write ONE concise paragraph (2–4 sentences) summarizing the anomalous event in the video. Your description must be consistent with the given category label and grounded in visible evidence.\n\n\n            Your paragraph must cover:\n\n                Scene context: describe the road setting, and mention time of day, weather, or road surface only if visible.\n\n                Object grounding: identify the main road user as object A, including type and color if visible, and describe where it is in the frame and how it moves.\n\n                Description: explain the main action or anomaly using clear driving terms.\n\n                Reasoning: briefly explain why the behavior is abnormal or unsafe based on visible evidence.\n            "
+  }
+]

roundabout_vau_test_videos.txt

@@ -0,0 +1,42 @@
+Car-Bike-Rangeline-CarmelDr-10-16-25.mp4
+Car-Car-L1-Towne-106TH-02-05-25.mp4
+Car-Car-L2-Springmill-Main-08-09-24.mp4
+Car-Car-PE-Gray-96TH-10-31-25.mp4
+Car-Car-PE-HazelDell-116TH-1-28-26.mp4
+Car-Car-PE-Rangeline-CityCenter-11-01-24.mp4
+Car-Rndbt-Cnstrct-Towne-Main-07-17-25.mp4
+Car-Rndbt-Rangeline-136TH-1-2-26.mp4
+Car-Rndbt-Springmill-106TH-4-12-24.mp4
+CarR3.mp4
+CarStall.mov
+Motorcycle-Rndbt-Rangeline-Elm-2020.mp4
+N-HazelDell-116TH-C2-Night_537-1008.mp4
+N-HazelDell-Main-C1-Day-Cloudy_6856-7175.mp4
+N-HazelDell-Main-C2-Night_6881-6981.mp4
+N-Illinois-Main-C2-Day-Sunny_2628-3261.mp4
+N-Springmill-Main-C2-Day-Sunny_3533-3926.mp4
+RangelineS116thSt_av_168_18_3.mp4
+RangelineS116thSt_av_216_27_1.mp4
+RangelineS116thSt_av_217_21_3.mp4
+RangelineS116thSt_av_221_19_4.mp4
+RangelineS116thSt_av_249_23_4.mp4
+RangelineS116thSt_av_303_18_3.mp4
+RangelineS116thSt_av_312_18_3.mp4
+RangelineS116thSt_av_364_21_3.mp4
+RangelineSMedicalDr_av_142_26_2.mp4
+RangelineSMedicalDr_av_180_18_3.mp4
+RangelineSMedicalDr_av_27_30_2.mp4
+RangelineSMedicalDr_av_296_21_3.mp4
+RangelineSMedicalDr_av_36_18_3.mp4
+RangelineSMedicalDr_av_43_18_3.mp4
+Trailer-Fall-HazelDell-116TH-12-04-23.mp4
+Truck-RT-Rangeline-Elm-10-21-25.mp4
+Truck-TR-Rangeline-Elm-2.mp4
+frames-000000_s3859_n421.mp4
+frames-000042_s2243_n364.mp4
+frames-000046_s4898_n538.mp4
+no11_cat23_sev4.mp4
+no21_cat23_sev4.mp4
+no22_cat22_sev4.mp4
+no4_cat23_sev4.mp4
+no6_cat23_sev4.mp4

roundabout_vau_time_window_qa_pairs.json

@@ -0,0 +1,3590 @@
+[
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 14]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_117_22_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 7]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_142_26_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 7]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_224_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 5]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_253_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 3]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_27_30_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 2]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_341_19_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0.5, 2]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_176_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_186_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0.8, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_198_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_305_14_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 11]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_53_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 8]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Gray-96TH-10-31-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[14, 16]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-HazelDell-116TH-09-13-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10.5, 12]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Keystone-116TH-09-03-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[18, 21]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Keystone-116TH-08-18-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[13, 16.5]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-DUI-Keystone-106TH-07-19-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6.5, 9]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Shelborne-126TH-11-04-23-Short.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 7]"
+      }
+    ],
+    "video": "/data/Motorcycle-Rndbt-Rangeline-Elm-2020.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[18, 20]"
+      }
+    ],
+    "video": "/data/Truck-RT-Rangeline-Elm-10-21-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[13, 15]"
+      }
+    ],
+    "video": "/data/Car-Car-NoStop-CityCenter-CarmelDr-12-20-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 9]"
+      }
+    ],
+    "video": "/data/no8_cat22_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 6]"
+      }
+    ],
+    "video": "/data/no24_cat13_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 10]"
+      }
+    ],
+    "video": "/data/no16_cat23_sev4_rm.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 2]"
+      }
+    ],
+    "video": "/data/no16_cat23_sev4_rc.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 10]"
+      }
+    ],
+    "video": "/data/no6_cat5_sev2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 4]"
+      }
+    ],
+    "video": "/data/no5_cat19_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 4]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_116_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1.5, 3]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_127_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 8]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_132_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 5]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_133_21_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 6]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_140_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 4]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_160_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 5]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_162_30_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 2]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_180_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[15, 19]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_181_27_1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 3]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_189_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 10]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_195_27_1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 6]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_200_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 11]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_211_14_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 6]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_216_27_1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 4]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_218_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 6]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_230_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 7.5]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_236_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 9]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_23_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 10]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_240_28_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 7]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_242_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 6]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_250_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[16, 20]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_252_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 5]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_253_30_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 5]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_257_19_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[18, 20]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_258_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 5]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_263_21_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 5]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_263_22_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 10]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_268_27_1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 4]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_273_30_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 8]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_278_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1.5, 7]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_279_22_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 11]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_280_30_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[12, 19]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_283_27_1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 14]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_289_27_1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 6]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_28_5_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 7]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_293_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 6]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_296_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 9]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_29_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 10]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_312_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1.5, 3]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_31_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 8]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_338_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 8]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_35_5_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 10]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_36_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 12]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_37_19_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 10]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_38_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 8]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_43_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 12]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_44_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[12, 16]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_65_27_1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[13, 15]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_80_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 11]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_93_27_1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 12]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_98_27_1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 13]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_111_19_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_114_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_116_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 14]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_121_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 2.5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_12_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 4]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_142_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 6]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_153_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[12, 16]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_168_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 20]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_169_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_16_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 11]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_17_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10, 12]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_183_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_188_14_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[33, 37]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_189_32_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 9]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_201_5_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_216_27_1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 4]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_217_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_221_19_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10, 12]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_222_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1.5, 3.5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_233_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[18, 21]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_235_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10, 12]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_237_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 10]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_244_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 2]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_249_23_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_253_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_270_30_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 12]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_276_23_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_277_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 4]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_27_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 10]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_280_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 11]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_281_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_286_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1.5, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_287_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2.0, 4.0]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_289_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 4]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_294_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 8]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_29_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 12]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_303_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0.8, 1.5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_308_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 3.5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_312_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_314_27_1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 10]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_315_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 12]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_319_19_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 4]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_321_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 6]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_322_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 4]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_324_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 2.5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_32_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 8]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_336_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[12, 16]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_337_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1.5, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_340_19_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10, 12]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_346_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 10]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_353_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[13, 15]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_364_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 13]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_37_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 4]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_38_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 4]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_39_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 9]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_40_5_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 2.5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_41_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1.5, 2.5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_48_19_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0.5, 3.5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_4_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 10]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_57_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 6]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_62_23_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1.5, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_64_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 9]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_73_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_7_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 3.5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_80_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 4]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_87_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 10]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_92_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[18, 100]"
+      }
+    ],
+    "video": "/data/3Truck-TR-Rangeline-Elm-12-6-22.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 9]"
+      }
+    ],
+    "video": "/data/Bus-Car-PE-Rangeline-Smoky-10-16-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[24, 26]"
+      }
+    ],
+    "video": "/data/Car-Bike-Keystone-Smoky-08-15-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[27, 29]"
+      }
+    ],
+    "video": "/data/Car-Bike-Rangeline-CarmelDr-10-16-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 6]"
+      }
+    ],
+    "video": "/data/Car-Car-L1-CityCenter-CarmelDr-07-29-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10.5, 12]"
+      }
+    ],
+    "video": "/data/Car-Car-L1-Illinois-Main-09-11-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[11, 16]"
+      }
+    ],
+    "video": "/data/Car-Car-L1-Rangeline-116TH-01-14-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 8]"
+      }
+    ],
+    "video": "/data/Car-Car-L1-Rangeline-116TH-10-26-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 9]"
+      }
+    ],
+    "video": "/data/Car-Car-L1-Rangeline-CarmelDr-11-03-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7.5, 10]"
+      }
+    ],
+    "video": "/data/Car-Car-L1-Towne-106TH-02-05-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 10]"
+      }
+    ],
+    "video": "/data/Car-Car-L2-Springmill-Main-08-09-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 11]"
+      }
+    ],
+    "video": "/data/Car-Car-LC-Keystone-Main-11-18-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[14, 17]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-3rd-CityCenter-02-10-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 10]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Gray-96TH-08-14-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[12, 15]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-HazelDell-116TH-02-22-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[51, 54]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Keystone-116TH-11-09-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10.5, 12]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Keystone-116TH-11-17-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 5]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Rangeline-116TH-03-03-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[14, 16]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Rangeline-116TH-06-18-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 5]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Rangeline-CityCenter-11-01-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 10]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Rangeline-Medical-08-15-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 12]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-RE-Rangeline-Medical-07-09-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10, 12]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-HazelDell-Main-9-26-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[16, 19]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Illinois-Main-03-05-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 12]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Illinois-Main-07-07-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 10]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Illinois-Main-10-23-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 11]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-PriorityWay-96TH-07-22-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[12, 14]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Rangeline-116TH-07-03-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[12, 15]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Rangeline-CarmelDr-03-10-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 6]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Rangeline-CarmelDr-03-12-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 11.5]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Rangeline-CarmelDr-05-22-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10, 13]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Rangeline-CarmelDr-06-28-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 12]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Rangeline-CarmelDr-10-14-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[11, 13]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-RE-HazelDell-116TH-11-18-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 10]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Towne-106TH-02-05-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10, 13]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Towne-106TH-10-08-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 11]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Towne-116TH-08-07-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[23, 25]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Towne-116TH-10-15-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10.5, 12]"
+      }
+    ],
+    "video": "/data/Car-Car-RR-SS-HazelDell-126TH-10-5-23.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[11, 13]"
+      }
+    ],
+    "video": "/data/Car-Car-Speed-PE-Rangeline-Medical-06-26-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 12]"
+      }
+    ],
+    "video": "/data/CarPoorEntrance.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 4]"
+      }
+    ],
+    "video": "/data/CarR3Gray.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 12]"
+      }
+    ],
+    "video": "/data/CarR3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[13, 15]"
+      }
+    ],
+    "video": "/data/CarR3Zotec.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 4]"
+      }
+    ],
+    "video": "/data/CarRearEnd.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10, 11]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Cnstrct-Towne-Main-07-17-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 35]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-DUI-WW-Springmill-Main-12-14-23.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4.5, 7]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-HazelDell-126TH-2023.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 31]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Keystone-106th-10-31-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 10]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Keystone-116TH-08-28-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 7]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Keystone-Smoky-8-23-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 10]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Main-4th-2023.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 10]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Rangeline-Smoky-12-01-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 14]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Shelborne-126TH-8-13-23-Short.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1.5, 3]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Smoky-Rangeline-2022.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 35]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Springmill-106TH-08-14-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 36]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Springmill-106TH-4-12-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 6]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Towne-116TH-02-10-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 41]"
+      }
+    ],
+    "video": "/data/CarStall.mov"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 22]"
+      }
+    ],
+    "video": "/data/CarWrongWayEntranceAndExit.mov"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 5]"
+      }
+    ],
+    "video": "/data/CarWrongWayEntrance.mov"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 18]"
+      }
+    ],
+    "video": "/data/CarWrongWay.mov"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 7]"
+      }
+    ],
+    "video": "/data/DriveIntoMedian.mov"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10, 14]"
+      }
+    ],
+    "video": "/data/ExitWrongWay.mov"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[13, 16]"
+      }
+    ],
+    "video": "/data/Trailer-Fall-HazelDell-116TH-12-04-23.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[17, 20]"
+      }
+    ],
+    "video": "/data/Truck-Car-L1-Rangeline-CarmelDr-10-22-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 7]"
+      }
+    ],
+    "video": "/data/Truck-Car-R3-Rangeline-116TH-07-16-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 6]"
+      }
+    ],
+    "video": "/data/TruckDropLoad.mov"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[15, 20]"
+      }
+    ],
+    "video": "/data/Truck-TR-Fall-Springmill-Main-05-13-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[62, 67]"
+      }
+    ],
+    "video": "/data/Truck-TR-Main-4th-10-25-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[15, 50]"
+      }
+    ],
+    "video": "/data/Truck-TR-Rangeline-Elm-08-17-2022.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[12, 25]"
+      }
+    ],
+    "video": "/data/Truck-TR-Rangeline-Elm-1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[56, 88]"
+      }
+    ],
+    "video": "/data/Truck-TR-Rangeline-Elm-2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[14, 48]"
+      }
+    ],
+    "video": "/data/Truck-TR-Rangline-Elm-10-24-19.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[20, 23]"
+      }
+    ],
+    "video": "/data/Truck-Truck-L1-Gray-116TH-10-22-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10, 119]"
+      }
+    ],
+    "video": "/data/Truck-TR-WW-Rangline-Elm-10-4-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 9]"
+      }
+    ],
+    "video": "/data/WrongWayUTurn.mov"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 7]"
+      }
+    ],
+    "video": "/data/Car-Car-L1-Rangeline-116TH-12-17-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 9]"
+      }
+    ],
+    "video": "/data/Car-Car-L1-Rangeline-CarmelDr-1-14-26.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10, 12]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-HazelDell-116TH-1-28-26.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10, 13]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-HazelDell-Main-1-21-26.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 11]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Keystone-116TH-12-30-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[14, 17]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Rangeline-116TH-12-29-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[13, 16]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Rangeline-CarmelDr-1-16-26-Cut.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 11]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-RearEnd-Rangeline-CarmelDr-12-19-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4.5, 6]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Gray-96TH-1-14-26.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 9]"
+      }
+    ],
+    "video": "/data/Car-Car-RearEnd-Gray-116TH-1-28-26.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 10]"
+      }
+    ],
+    "video": "/data/Car-Car-RearEnd-Keystone-116TH-1-4-26.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 28]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-DelegatesRow-96TH-1-5-26.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 9]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Rangeline-136TH-1-2-26.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 7]"
+      }
+    ],
+    "video": "/data/no26_cat32_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 10]"
+      }
+    ],
+    "video": "/data/no10_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 10]"
+      }
+    ],
+    "video": "/data/no4_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 10]"
+      }
+    ],
+    "video": "/data/no3_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 11]"
+      }
+    ],
+    "video": "/data/no15_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 6]"
+      }
+    ],
+    "video": "/data/no20_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 4]"
+      }
+    ],
+    "video": "/data/no11_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 10]"
+      }
+    ],
+    "video": "/data/no21_cat13_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 9]"
+      }
+    ],
+    "video": "/data/no17_cat6_sev1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 6]"
+      }
+    ],
+    "video": "/data/no23_cat18_sev3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 4]"
+      }
+    ],
+    "video": "/data/no18_cat21_sev3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 10]"
+      }
+    ],
+    "video": "/data/no29_cat21_sev3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 9]"
+      }
+    ],
+    "video": "/data/no6_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 10]"
+      }
+    ],
+    "video": "/data/no27_cat21_sev3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 5]"
+      }
+    ],
+    "video": "/data/no22_cat22_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 9]"
+      }
+    ],
+    "video": "/data/no5_cat5_sev2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 9]"
+      }
+    ],
+    "video": "/data/no1_cat3_sev1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 10]"
+      }
+    ],
+    "video": "/data/no25_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 8]"
+      }
+    ],
+    "video": "/data/no9_cat22_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 6]"
+      }
+    ],
+    "video": "/data/no13_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 6]"
+      }
+    ],
+    "video": "/data/no7_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 6]"
+      }
+    ],
+    "video": "/data/no2_cat18_sev3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 9]"
+      }
+    ],
+    "video": "/data/no14_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 10]"
+      }
+    ],
+    "video": "/data/no28_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4.5, 9]"
+      }
+    ],
+    "video": "/data/no7_cat29_sev3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 5]"
+      }
+    ],
+    "video": "/data/no10_cat31_sev2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1.5, 7]"
+      }
+    ],
+    "video": "/data/no14_cat29_sev3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 10]"
+      }
+    ],
+    "video": "/data/no18_cat9_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 10]"
+      }
+    ],
+    "video": "/data/no15_cat14_sev3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 9]"
+      }
+    ],
+    "video": "/data/no20_cat14_sev3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 3]"
+      }
+    ],
+    "video": "/data/no8_cat21_sev3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 9]"
+      }
+    ],
+    "video": "/data/no13_cat27_sev1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 10]"
+      }
+    ],
+    "video": "/data/no11_cat4_sev1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 8]"
+      }
+    ],
+    "video": "/data/no1_cat19_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 9]"
+      }
+    ],
+    "video": "/data/no4_cat32_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 10]"
+      }
+    ],
+    "video": "/data/no21_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 9]"
+      }
+    ],
+    "video": "/data/no22_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 7]"
+      }
+    ],
+    "video": "/data/no12_cat26_sev2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 10]"
+      }
+    ],
+    "video": "/data/no3_cat22_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 10]"
+      }
+    ],
+    "video": "/data/no19_cat25_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 5]"
+      }
+    ],
+    "video": "/data/no9_cat14_sev3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 10]"
+      }
+    ],
+    "video": "/data/no2_cat23_sev4.mp4"
+  }
+]

roundabout_vau_time_window_qa_pairs_test.json

@@ -0,0 +1,444 @@
+[
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 7]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_142_26_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 3]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_27_30_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 8]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Gray-96TH-10-31-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 7]"
+      }
+    ],
+    "video": "/data/Motorcycle-Rndbt-Rangeline-Elm-2020.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[18, 20]"
+      }
+    ],
+    "video": "/data/Truck-RT-Rangeline-Elm-10-21-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 2]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_180_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 6]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_296_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 10]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_36_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 8]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_43_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[12, 16]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_168_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_216_27_1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 4]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_217_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_221_19_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 2]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_249_23_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 12]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_303_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 3.5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_312_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[13, 15]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_364_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[27, 29]"
+      }
+    ],
+    "video": "/data/Car-Bike-Rangeline-CarmelDr-10-16-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7.5, 10]"
+      }
+    ],
+    "video": "/data/Car-Car-L1-Towne-106TH-02-05-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 10]"
+      }
+    ],
+    "video": "/data/Car-Car-L2-Springmill-Main-08-09-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 5]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Rangeline-CityCenter-11-01-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 12]"
+      }
+    ],
+    "video": "/data/CarR3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10, 11]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Cnstrct-Towne-Main-07-17-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 36]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Springmill-106TH-4-12-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 41]"
+      }
+    ],
+    "video": "/data/CarStall.mov"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[13, 16]"
+      }
+    ],
+    "video": "/data/Trailer-Fall-HazelDell-116TH-12-04-23.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[56, 88]"
+      }
+    ],
+    "video": "/data/Truck-TR-Rangeline-Elm-2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10, 12]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-HazelDell-116TH-1-28-26.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 9]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Rangeline-136TH-1-2-26.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 10]"
+      }
+    ],
+    "video": "/data/no4_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 4]"
+      }
+    ],
+    "video": "/data/no11_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 9]"
+      }
+    ],
+    "video": "/data/no6_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 5]"
+      }
+    ],
+    "video": "/data/no22_cat22_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 10]"
+      }
+    ],
+    "video": "/data/no21_cat23_sev4.mp4"
+  }
+]

roundabout_vau_time_window_qa_pairs_train.json

@@ -0,0 +1,3148 @@
+[
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 14]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_117_22_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 7]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_224_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 5]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_253_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 2]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_341_19_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0.5, 2]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_176_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_186_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0.8, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_198_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_305_14_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 11]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_53_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[14, 16]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-HazelDell-116TH-09-13-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10.5, 12]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Keystone-116TH-09-03-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[18, 21]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Keystone-116TH-08-18-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[13, 16.5]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-DUI-Keystone-106TH-07-19-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6.5, 9]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Shelborne-126TH-11-04-23-Short.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[13, 15]"
+      }
+    ],
+    "video": "/data/Car-Car-NoStop-CityCenter-CarmelDr-12-20-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 9]"
+      }
+    ],
+    "video": "/data/no8_cat22_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 6]"
+      }
+    ],
+    "video": "/data/no24_cat13_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 10]"
+      }
+    ],
+    "video": "/data/no16_cat23_sev4_rm.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 2]"
+      }
+    ],
+    "video": "/data/no16_cat23_sev4_rc.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 10]"
+      }
+    ],
+    "video": "/data/no6_cat5_sev2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 4]"
+      }
+    ],
+    "video": "/data/no5_cat19_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 4]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_116_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1.5, 3]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_127_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 8]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_132_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 5]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_133_21_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 6]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_140_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 4]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_160_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 5]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_162_30_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[15, 19]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_181_27_1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 3]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_189_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 10]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_195_27_1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 6]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_200_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 11]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_211_14_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 6]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_216_27_1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 4]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_218_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 6]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_230_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 7.5]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_236_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 9]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_23_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 10]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_240_28_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 7]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_242_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 6]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_250_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[16, 20]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_252_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 5]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_253_30_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 5]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_257_19_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[18, 20]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_258_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 5]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_263_21_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 5]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_263_22_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 10]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_268_27_1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 4]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_273_30_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 8]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_278_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1.5, 7]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_279_22_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 11]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_280_30_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[12, 19]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_283_27_1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 14]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_289_27_1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 6]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_28_5_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 7]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_293_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 9]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_29_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 10]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_312_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1.5, 3]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_31_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 8]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_338_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 8]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_35_5_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 12]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_37_19_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 10]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_38_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 12]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_44_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[12, 16]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_65_27_1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[13, 15]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_80_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 11]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_93_27_1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 12]"
+      }
+    ],
+    "video": "/data/RangelineSMedicalDr_av_98_27_1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 13]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_111_19_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_114_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_116_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 14]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_121_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 2.5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_12_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 4]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_142_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 6]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_153_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 20]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_169_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_16_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 11]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_17_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10, 12]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_183_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_188_14_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[33, 37]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_189_32_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 9]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_201_5_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10, 12]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_222_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1.5, 3.5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_233_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[18, 21]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_235_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10, 12]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_237_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 10]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_244_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_253_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_270_30_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 12]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_276_23_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_277_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 4]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_27_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 10]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_280_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 11]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_281_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_286_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1.5, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_287_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2.0, 4.0]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_289_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 4]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_294_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 8]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_29_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0.8, 1.5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_308_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_314_27_1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 10]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_315_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 12]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_319_19_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 4]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_321_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 6]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_322_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 4]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_324_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 2.5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_32_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 8]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_336_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[12, 16]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_337_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1.5, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_340_19_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10, 12]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_346_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 10]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_353_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 13]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_37_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 4]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_38_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 4]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_39_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 9]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_40_5_2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 2.5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_41_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1.5, 2.5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_48_19_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0.5, 3.5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_4_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 10]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_57_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 6]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_62_23_4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1.5, 3]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_64_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 9]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_73_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_7_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 3.5]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_80_18_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 4]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_87_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 10]"
+      }
+    ],
+    "video": "/data/RangelineS116thSt_av_92_21_3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[18, 100]"
+      }
+    ],
+    "video": "/data/3Truck-TR-Rangeline-Elm-12-6-22.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 9]"
+      }
+    ],
+    "video": "/data/Bus-Car-PE-Rangeline-Smoky-10-16-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[24, 26]"
+      }
+    ],
+    "video": "/data/Car-Bike-Keystone-Smoky-08-15-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 6]"
+      }
+    ],
+    "video": "/data/Car-Car-L1-CityCenter-CarmelDr-07-29-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10.5, 12]"
+      }
+    ],
+    "video": "/data/Car-Car-L1-Illinois-Main-09-11-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[11, 16]"
+      }
+    ],
+    "video": "/data/Car-Car-L1-Rangeline-116TH-01-14-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 8]"
+      }
+    ],
+    "video": "/data/Car-Car-L1-Rangeline-116TH-10-26-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 9]"
+      }
+    ],
+    "video": "/data/Car-Car-L1-Rangeline-CarmelDr-11-03-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 11]"
+      }
+    ],
+    "video": "/data/Car-Car-LC-Keystone-Main-11-18-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[14, 17]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-3rd-CityCenter-02-10-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 10]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Gray-96TH-08-14-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[12, 15]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-HazelDell-116TH-02-22-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[51, 54]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Keystone-116TH-11-09-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10.5, 12]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Keystone-116TH-11-17-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 5]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Rangeline-116TH-03-03-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[14, 16]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Rangeline-116TH-06-18-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 10]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Rangeline-Medical-08-15-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 12]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-RE-Rangeline-Medical-07-09-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10, 12]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-HazelDell-Main-9-26-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[16, 19]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Illinois-Main-03-05-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 12]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Illinois-Main-07-07-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 10]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Illinois-Main-10-23-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 11]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-PriorityWay-96TH-07-22-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[12, 14]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Rangeline-116TH-07-03-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[12, 15]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Rangeline-CarmelDr-03-10-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 6]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Rangeline-CarmelDr-03-12-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 11.5]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Rangeline-CarmelDr-05-22-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10, 13]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Rangeline-CarmelDr-06-28-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 12]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Rangeline-CarmelDr-10-14-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[11, 13]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-RE-HazelDell-116TH-11-18-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 10]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Towne-106TH-02-05-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10, 13]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Towne-106TH-10-08-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 11]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Towne-116TH-08-07-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[23, 25]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Towne-116TH-10-15-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10.5, 12]"
+      }
+    ],
+    "video": "/data/Car-Car-RR-SS-HazelDell-126TH-10-5-23.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[11, 13]"
+      }
+    ],
+    "video": "/data/Car-Car-Speed-PE-Rangeline-Medical-06-26-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 12]"
+      }
+    ],
+    "video": "/data/CarPoorEntrance.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 4]"
+      }
+    ],
+    "video": "/data/CarR3Gray.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[13, 15]"
+      }
+    ],
+    "video": "/data/CarR3Zotec.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 4]"
+      }
+    ],
+    "video": "/data/CarRearEnd.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[8, 35]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-DUI-WW-Springmill-Main-12-14-23.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4.5, 7]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-HazelDell-126TH-2023.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 31]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Keystone-106th-10-31-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 10]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Keystone-116TH-08-28-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 7]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Keystone-Smoky-8-23-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 10]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Main-4th-2023.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 10]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Rangeline-Smoky-12-01-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 14]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Shelborne-126TH-8-13-23-Short.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1.5, 3]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Smoky-Rangeline-2022.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 35]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Springmill-106TH-08-14-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 6]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-Towne-116TH-02-10-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 22]"
+      }
+    ],
+    "video": "/data/CarWrongWayEntranceAndExit.mov"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 5]"
+      }
+    ],
+    "video": "/data/CarWrongWayEntrance.mov"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 18]"
+      }
+    ],
+    "video": "/data/CarWrongWay.mov"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 7]"
+      }
+    ],
+    "video": "/data/DriveIntoMedian.mov"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10, 14]"
+      }
+    ],
+    "video": "/data/ExitWrongWay.mov"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[17, 20]"
+      }
+    ],
+    "video": "/data/Truck-Car-L1-Rangeline-CarmelDr-10-22-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 7]"
+      }
+    ],
+    "video": "/data/Truck-Car-R3-Rangeline-116TH-07-16-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 6]"
+      }
+    ],
+    "video": "/data/TruckDropLoad.mov"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[15, 20]"
+      }
+    ],
+    "video": "/data/Truck-TR-Fall-Springmill-Main-05-13-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[62, 67]"
+      }
+    ],
+    "video": "/data/Truck-TR-Main-4th-10-25-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[15, 50]"
+      }
+    ],
+    "video": "/data/Truck-TR-Rangeline-Elm-08-17-2022.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[12, 25]"
+      }
+    ],
+    "video": "/data/Truck-TR-Rangeline-Elm-1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[14, 48]"
+      }
+    ],
+    "video": "/data/Truck-TR-Rangline-Elm-10-24-19.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[20, 23]"
+      }
+    ],
+    "video": "/data/Truck-Truck-L1-Gray-116TH-10-22-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10, 119]"
+      }
+    ],
+    "video": "/data/Truck-TR-WW-Rangline-Elm-10-4-24.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 9]"
+      }
+    ],
+    "video": "/data/WrongWayUTurn.mov"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 7]"
+      }
+    ],
+    "video": "/data/Car-Car-L1-Rangeline-116TH-12-17-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[7, 9]"
+      }
+    ],
+    "video": "/data/Car-Car-L1-Rangeline-CarmelDr-1-14-26.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[10, 13]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-HazelDell-Main-1-21-26.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 11]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Keystone-116TH-12-30-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[14, 17]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Rangeline-116TH-12-29-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[13, 16]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-Rangeline-CarmelDr-1-16-26-Cut.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[9, 11]"
+      }
+    ],
+    "video": "/data/Car-Car-PE-RearEnd-Rangeline-CarmelDr-12-19-25.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4.5, 6]"
+      }
+    ],
+    "video": "/data/Car-Car-R3-Gray-96TH-1-14-26.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 9]"
+      }
+    ],
+    "video": "/data/Car-Car-RearEnd-Gray-116TH-1-28-26.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 10]"
+      }
+    ],
+    "video": "/data/Car-Car-RearEnd-Keystone-116TH-1-4-26.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 28]"
+      }
+    ],
+    "video": "/data/Car-Rndbt-DelegatesRow-96TH-1-5-26.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 7]"
+      }
+    ],
+    "video": "/data/no26_cat32_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 10]"
+      }
+    ],
+    "video": "/data/no10_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 10]"
+      }
+    ],
+    "video": "/data/no3_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 11]"
+      }
+    ],
+    "video": "/data/no15_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 6]"
+      }
+    ],
+    "video": "/data/no20_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 10]"
+      }
+    ],
+    "video": "/data/no21_cat13_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1, 9]"
+      }
+    ],
+    "video": "/data/no17_cat6_sev1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 6]"
+      }
+    ],
+    "video": "/data/no23_cat18_sev3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 4]"
+      }
+    ],
+    "video": "/data/no18_cat21_sev3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 10]"
+      }
+    ],
+    "video": "/data/no29_cat21_sev3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 10]"
+      }
+    ],
+    "video": "/data/no27_cat21_sev3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 9]"
+      }
+    ],
+    "video": "/data/no5_cat5_sev2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 9]"
+      }
+    ],
+    "video": "/data/no1_cat3_sev1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 10]"
+      }
+    ],
+    "video": "/data/no25_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4, 8]"
+      }
+    ],
+    "video": "/data/no9_cat22_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 6]"
+      }
+    ],
+    "video": "/data/no13_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 6]"
+      }
+    ],
+    "video": "/data/no7_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 6]"
+      }
+    ],
+    "video": "/data/no2_cat18_sev3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 9]"
+      }
+    ],
+    "video": "/data/no14_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[5, 10]"
+      }
+    ],
+    "video": "/data/no28_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[4.5, 9]"
+      }
+    ],
+    "video": "/data/no7_cat29_sev3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 5]"
+      }
+    ],
+    "video": "/data/no10_cat31_sev2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[1.5, 7]"
+      }
+    ],
+    "video": "/data/no14_cat29_sev3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 10]"
+      }
+    ],
+    "video": "/data/no18_cat9_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 10]"
+      }
+    ],
+    "video": "/data/no15_cat14_sev3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 9]"
+      }
+    ],
+    "video": "/data/no20_cat14_sev3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 3]"
+      }
+    ],
+    "video": "/data/no8_cat21_sev3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 9]"
+      }
+    ],
+    "video": "/data/no13_cat27_sev1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 10]"
+      }
+    ],
+    "video": "/data/no11_cat4_sev1.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[6, 8]"
+      }
+    ],
+    "video": "/data/no1_cat19_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 9]"
+      }
+    ],
+    "video": "/data/no4_cat32_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 9]"
+      }
+    ],
+    "video": "/data/no22_cat23_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 7]"
+      }
+    ],
+    "video": "/data/no12_cat26_sev2.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 10]"
+      }
+    ],
+    "video": "/data/no3_cat22_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[0, 10]"
+      }
+    ],
+    "video": "/data/no19_cat25_sev4.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[2, 5]"
+      }
+    ],
+    "video": "/data/no9_cat14_sev3.mp4"
+  },
+  {
+    "conversations": [
+      {
+        "from": "human",
+        "value": "<video>\nInspect the whole video and provide a time window [start_sec, end_sec] that contains the main anomaly event."
+      },
+      {
+        "from": "gpt",
+        "value": "[3, 10]"
+      }
+    ],
+    "video": "/data/no2_cat23_sev4.mp4"
+  }
+]

roundabout_vau_train_videos.txt

@@ -0,0 +1,300 @@
+3Truck-TR-Rangeline-Elm-12-6-22.mp4
+Bus-Car-PE-Rangeline-Smoky-10-16-25.mp4
+Car-Bike-Keystone-Smoky-08-15-24.mp4
+Car-Car-L1-CityCenter-CarmelDr-07-29-24.mp4
+Car-Car-L1-Illinois-Main-09-11-25.mp4
+Car-Car-L1-Rangeline-116TH-01-14-25.mp4
+Car-Car-L1-Rangeline-116TH-10-26-24.mp4
+Car-Car-L1-Rangeline-116TH-12-17-25.mp4
+Car-Car-L1-Rangeline-CarmelDr-1-14-26.mp4
+Car-Car-L1-Rangeline-CarmelDr-11-03-25.mp4
+Car-Car-LC-Keystone-Main-11-18-25.mp4
+Car-Car-NoStop-CityCenter-CarmelDr-12-20-25.mp4
+Car-Car-PE-3rd-CityCenter-02-10-25.mp4
+Car-Car-PE-Gray-96TH-08-14-24.mp4
+Car-Car-PE-HazelDell-116TH-02-22-25.mp4
+Car-Car-PE-HazelDell-116TH-09-13-24.mp4
+Car-Car-PE-HazelDell-Main-1-21-26.mp4
+Car-Car-PE-Keystone-116TH-09-03-25.mp4
+Car-Car-PE-Keystone-116TH-11-09-25.mp4
+Car-Car-PE-Keystone-116TH-11-17-25.mp4
+Car-Car-PE-Keystone-116TH-12-30-25.mp4
+Car-Car-PE-RE-Rangeline-Medical-07-09-25.mp4
+Car-Car-PE-Rangeline-116TH-03-03-24.mp4
+Car-Car-PE-Rangeline-116TH-06-18-24.mp4
+Car-Car-PE-Rangeline-116TH-12-29-25.mp4
+Car-Car-PE-Rangeline-CarmelDr-1-16-26-Cut.mp4
+Car-Car-PE-Rangeline-Medical-08-15-24.mp4
+Car-Car-PE-RearEnd-Rangeline-CarmelDr-12-19-25.mp4
+Car-Car-R3-Gray-96TH-1-14-26.mp4
+Car-Car-R3-HazelDell-Main-9-26-24.mp4
+Car-Car-R3-Illinois-Main-03-05-25.mp4
+Car-Car-R3-Illinois-Main-07-07-25.mp4
+Car-Car-R3-Illinois-Main-10-23-25.mp4
+Car-Car-R3-Keystone-116TH-08-18-25.mp4
+Car-Car-R3-PriorityWay-96TH-07-22-25.mp4
+Car-Car-R3-RE-HazelDell-116TH-11-18-25.mp4
+Car-Car-R3-Rangeline-116TH-07-03-25.mp4
+Car-Car-R3-Rangeline-CarmelDr-03-10-25.mp4
+Car-Car-R3-Rangeline-CarmelDr-03-12-25.mp4
+Car-Car-R3-Rangeline-CarmelDr-05-22-24.mp4
+Car-Car-R3-Rangeline-CarmelDr-06-28-24.mp4
+Car-Car-R3-Rangeline-CarmelDr-10-14-25.mp4
+Car-Car-R3-Towne-106TH-02-05-25.mp4
+Car-Car-R3-Towne-106TH-10-08-24.mp4
+Car-Car-R3-Towne-116TH-08-07-24.mp4
+Car-Car-R3-Towne-116TH-10-15-24.mp4
+Car-Car-RR-SS-HazelDell-126TH-10-5-23.mp4
+Car-Car-RearEnd-Gray-116TH-1-28-26.mp4
+Car-Car-RearEnd-Keystone-116TH-1-4-26.mp4
+Car-Car-Speed-PE-Rangeline-Medical-06-26-24.mp4
+Car-Rndbt-DUI-Keystone-106TH-07-19-25.mp4
+Car-Rndbt-DUI-WW-Springmill-Main-12-14-23.mp4
+Car-Rndbt-DelegatesRow-96TH-1-5-26.mp4
+Car-Rndbt-HazelDell-126TH-2023.mp4
+Car-Rndbt-Keystone-106th-10-31-25.mp4
+Car-Rndbt-Keystone-116TH-08-28-25.mp4
+Car-Rndbt-Keystone-Smoky-8-23-25.mp4
+Car-Rndbt-Main-4th-2023.mp4
+Car-Rndbt-Rangeline-Smoky-12-01-25.mp4
+Car-Rndbt-Shelborne-126TH-11-04-23-Short.mp4
+Car-Rndbt-Shelborne-126TH-8-13-23-Short.mp4
+Car-Rndbt-Smoky-Rangeline-2022.mp4
+Car-Rndbt-Springmill-106TH-08-14-25.mp4
+Car-Rndbt-Towne-116TH-02-10-25.mp4
+CarPoorEntrance.mp4
+CarR3Gray.mp4
+CarR3Zotec.mp4
+CarRearEnd.mp4
+CarWrongWay.mov
+CarWrongWayEntrance.mov
+CarWrongWayEntranceAndExit.mov
+DriveIntoMedian.mov
+ExitWrongWay.mov
+N-CityCenter-3rdAV-C1-Night_7899-8419.mp4
+N-CityCenter-3rdAV-C2-Day-Cloudy_6665-7298.mp4
+N-CityCenter-CarmelDr-C1-Day-Cloudy_7878-8249.mp4
+N-CityCenter-CarmelDr-C2-Morning_395-555.mp4
+N-DelegatesRow-96TH-C1-Morning_5653-6176.mp4
+N-DelegatesRow-96TH-C2-Day-Sunny_1020-1519.mp4
+N-Gray-116TH-C1-Night_6362-6759.mp4
+N-Gray-116TH-C2-Morning-Sunrise_7428-7617.mp4
+N-Gray-96TH-C1-Night-Wet_8344-8547.mp4
+N-Gray-96TH-C2-Day-Cloudy_201-721.mp4
+N-HazelDell-116TH-C1-Day-Sunny_4602-5349.mp4
+N-HazelDell-126TH-C1-Day_5385-5493.mp4
+N-HazelDell-126TH-C2-Morning_1310-1615.mp4
+N-Illinois-Main-C1-Morning_3305-3518.mp4
+N-Illinois-Zotec-C1-Morning_8334-9089.mp4
+N-Illinois-Zotec-C2-Day-Snow_7965-8298.mp4
+N-Keystone-106TH-C1-Morning_172-518.mp4
+N-Keystone-106TH-C2-Day-Sunny_6-419.mp4
+N-Keystone-116TH-C1-Day-Sunny_5498-6298.mp4
+N-Keystone-116TH-C2-Night_4551-4822.mp4
+N-Keystone-Main-C1-Sunrise_520-775.mp4
+N-Keystone-Main-C2-Day-Snowy_8354-8857.mp4
+N-Keystone-SmokyRow-C1-Day-Cloudy_2355-2977.mp4
+N-Keystone-SmokyRow-C2-Day-Sunny_1585-1852.mp4
+N-Main-4thAV-C1-Day-Sunny_7391-7508.mp4
+N-Main-4thAV-C2-Night_4610-5220.mp4
+N-PriorityWay-96TH-C1-Evening-Sunset_1060-1477.mp4
+N-PriorityWay-96TH-C2-Day-Sunny_2356-2515.mp4
+N-Rangeline-116TH-C1-Day-Cloudy_2870-3204.mp4
+N-Rangeline-116TH-C2-Day-Sunny_356-760.mp4
+N-Rangeline-CarmelDr-C1-Day-Cloudy_3140-3546.mp4
+N-Rangeline-CarmelDr-C2-Night_5706-6177.mp4
+N-Rangeline-CityCenter-C1-Night_1960-2380.mp4
+N-Rangeline-CityCenter-C2-Day-Snowy_4483-4786.mp4
+N-Rangeline-Elm-C1-Morning-Sunny_1551-2009.mp4
+N-Rangeline-Elm-C2-Night_3245-4024.mp4
+N-Rangeline-Medical-C1-Day-Rainy_7436-7720.mp4
+N-Rangeline-Medical-C2-Night_5062-5540.mp4
+N-Rangeline-SmokyRow-C1-Morning-Sunny_8188-8372.mp4
+N-Rangeline-SmokyRow-C2-Evening-Cloudy_7358-7464.mp4
+N-Shelbourne-126TH-C1-Day-Sunny_6967-7363.mp4
+N-Shelbourne-126TH-C2-Evening_7059-7634.mp4
+N-Springmill-106TH-C1-Day-Cloudy_1254-1938.mp4
+N-Springmill-106TH-C2-Morning-Rainy_1035-1474.mp4
+N-Springmill-Main-C1-Morning-Rainy_6602-6801.mp4
+N-Towne-106TH-C1-Day_4472-5251.mp4
+N-Towne-106TH-C2-Evening-Sunny_7671-8041.mp4
+N-Towne-116TH-C1-Night_7387-8044.mp4
+N-Towne-116TH-C2-Morning_8342-8484.mp4
+N-Towne-Main-C1-Night-Snowing_301-899.mp4
+N-Towne-Main-C2-Day-Cloud_4311-4715.mp4
+RangelineS116thSt_av_111_19_4.mp4
+RangelineS116thSt_av_114_21_3.mp4
+RangelineS116thSt_av_116_21_3.mp4
+RangelineS116thSt_av_121_21_3.mp4
+RangelineS116thSt_av_12_18_3.mp4
+RangelineS116thSt_av_142_21_3.mp4
+RangelineS116thSt_av_153_18_3.mp4
+RangelineS116thSt_av_169_18_3.mp4
+RangelineS116thSt_av_16_18_3.mp4
+RangelineS116thSt_av_176_21_3.mp4
+RangelineS116thSt_av_17_18_3.mp4
+RangelineS116thSt_av_183_18_3.mp4
+RangelineS116thSt_av_186_21_3.mp4
+RangelineS116thSt_av_188_14_3.mp4
+RangelineS116thSt_av_189_32_4.mp4
+RangelineS116thSt_av_198_18_3.mp4
+RangelineS116thSt_av_201_5_2.mp4
+RangelineS116thSt_av_222_18_3.mp4
+RangelineS116thSt_av_233_18_3.mp4
+RangelineS116thSt_av_235_21_3.mp4
+RangelineS116thSt_av_237_21_3.mp4
+RangelineS116thSt_av_244_21_3.mp4
+RangelineS116thSt_av_253_21_3.mp4
+RangelineS116thSt_av_270_30_2.mp4
+RangelineS116thSt_av_276_23_4.mp4
+RangelineS116thSt_av_277_18_3.mp4
+RangelineS116thSt_av_27_18_3.mp4
+RangelineS116thSt_av_280_21_3.mp4
+RangelineS116thSt_av_281_18_3.mp4
+RangelineS116thSt_av_286_18_3.mp4
+RangelineS116thSt_av_287_21_3.mp4
+RangelineS116thSt_av_289_18_3.mp4
+RangelineS116thSt_av_294_21_3.mp4
+RangelineS116thSt_av_29_21_3.mp4
+RangelineS116thSt_av_305_14_3.mp4
+RangelineS116thSt_av_308_18_3.mp4
+RangelineS116thSt_av_314_27_1.mp4
+RangelineS116thSt_av_315_21_3.mp4
+RangelineS116thSt_av_319_19_4.mp4
+RangelineS116thSt_av_321_18_3.mp4
+RangelineS116thSt_av_322_18_3.mp4
+RangelineS116thSt_av_324_18_3.mp4
+RangelineS116thSt_av_32_21_3.mp4
+RangelineS116thSt_av_336_18_3.mp4
+RangelineS116thSt_av_337_18_3.mp4
+RangelineS116thSt_av_340_19_4.mp4
+RangelineS116thSt_av_346_18_3.mp4
+RangelineS116thSt_av_353_18_3.mp4
+RangelineS116thSt_av_37_21_3.mp4
+RangelineS116thSt_av_38_18_3.mp4
+RangelineS116thSt_av_39_18_3.mp4
+RangelineS116thSt_av_40_5_2.mp4
+RangelineS116thSt_av_41_21_3.mp4
+RangelineS116thSt_av_48_19_4.mp4
+RangelineS116thSt_av_4_18_3.mp4
+RangelineS116thSt_av_53_18_3.mp4
+RangelineS116thSt_av_57_21_3.mp4
+RangelineS116thSt_av_62_23_4.mp4
+RangelineS116thSt_av_64_21_3.mp4
+RangelineS116thSt_av_73_21_3.mp4
+RangelineS116thSt_av_7_18_3.mp4
+RangelineS116thSt_av_80_18_3.mp4
+RangelineS116thSt_av_87_21_3.mp4
+RangelineS116thSt_av_92_21_3.mp4
+RangelineSMedicalDr_av_116_18_3.mp4
+RangelineSMedicalDr_av_117_22_4.mp4
+RangelineSMedicalDr_av_127_18_3.mp4
+RangelineSMedicalDr_av_132_21_3.mp4
+RangelineSMedicalDr_av_133_21_4.mp4
+RangelineSMedicalDr_av_140_21_3.mp4
+RangelineSMedicalDr_av_160_18_3.mp4
+RangelineSMedicalDr_av_162_30_2.mp4
+RangelineSMedicalDr_av_181_27_1.mp4
+RangelineSMedicalDr_av_189_18_3.mp4
+RangelineSMedicalDr_av_195_27_1.mp4
+RangelineSMedicalDr_av_200_21_3.mp4
+RangelineSMedicalDr_av_211_14_3.mp4
+RangelineSMedicalDr_av_216_27_1.mp4
+RangelineSMedicalDr_av_218_18_3.mp4
+RangelineSMedicalDr_av_224_21_3.mp4
+RangelineSMedicalDr_av_230_18_3.mp4
+RangelineSMedicalDr_av_236_18_3.mp4
+RangelineSMedicalDr_av_23_18_3.mp4
+RangelineSMedicalDr_av_240_28_3.mp4
+RangelineSMedicalDr_av_242_21_3.mp4
+RangelineSMedicalDr_av_250_21_3.mp4
+RangelineSMedicalDr_av_252_18_3.mp4
+RangelineSMedicalDr_av_253_18_3.mp4
+RangelineSMedicalDr_av_253_30_2.mp4
+RangelineSMedicalDr_av_257_19_4.mp4
+RangelineSMedicalDr_av_258_18_3.mp4
+RangelineSMedicalDr_av_263_21_4.mp4
+RangelineSMedicalDr_av_263_22_4.mp4
+RangelineSMedicalDr_av_268_27_1.mp4
+RangelineSMedicalDr_av_273_30_2.mp4
+RangelineSMedicalDr_av_278_18_3.mp4
+RangelineSMedicalDr_av_279_22_4.mp4
+RangelineSMedicalDr_av_280_30_2.mp4
+RangelineSMedicalDr_av_283_27_1.mp4
+RangelineSMedicalDr_av_289_27_1.mp4
+RangelineSMedicalDr_av_28_5_2.mp4
+RangelineSMedicalDr_av_293_18_3.mp4
+RangelineSMedicalDr_av_29_18_3.mp4
+RangelineSMedicalDr_av_312_18_3.mp4
+RangelineSMedicalDr_av_31_18_3.mp4
+RangelineSMedicalDr_av_338_21_3.mp4
+RangelineSMedicalDr_av_341_19_4.mp4
+RangelineSMedicalDr_av_35_5_2.mp4
+RangelineSMedicalDr_av_37_19_4.mp4
+RangelineSMedicalDr_av_38_18_3.mp4
+RangelineSMedicalDr_av_44_18_3.mp4
+RangelineSMedicalDr_av_65_27_1.mp4
+RangelineSMedicalDr_av_80_18_3.mp4
+RangelineSMedicalDr_av_93_27_1.mp4
+RangelineSMedicalDr_av_98_27_1.mp4
+Truck-Car-L1-Rangeline-CarmelDr-10-22-24.mp4
+Truck-Car-R3-Rangeline-116TH-07-16-25.mp4
+Truck-TR-Fall-Springmill-Main-05-13-24.mp4
+Truck-TR-Main-4th-10-25-25.mp4
+Truck-TR-Rangeline-Elm-08-17-2022.mp4
+Truck-TR-Rangeline-Elm-1.mp4
+Truck-TR-Rangline-Elm-10-24-19.mp4
+Truck-TR-WW-Rangline-Elm-10-4-24.mp4
+Truck-Truck-L1-Gray-116TH-10-22-25.mp4
+TruckDropLoad.mov
+WrongWayUTurn.mov
+frames-000000_s6296_n328.mp4
+frames-000002_s1526_n252.mp4
+frames-000015_s914_n165.mp4
+frames-000049_s1821_n429.mp4
+frames-000061_s3207_n134.mp4
+frames-000067_s5740_n453.mp4
+frames-000085_s5586_n93.mp4
+no10_cat23_sev4.mp4
+no10_cat31_sev2.mp4
+no11_cat4_sev1.mp4
+no12_cat26_sev2.mp4
+no13_cat23_sev4.mp4
+no13_cat27_sev1.mp4
+no14_cat23_sev4.mp4
+no14_cat29_sev3.mp4
+no15_cat14_sev3.mp4
+no15_cat23_sev4.mp4
+no16_cat23_sev4_rc.mp4
+no16_cat23_sev4_rm.mp4
+no17_cat6_sev1.mp4
+no18_cat21_sev3.mp4
+no18_cat9_sev4.mp4
+no19_cat25_sev4.mp4
+no1_cat19_sev4.mp4
+no1_cat3_sev1.mp4
+no20_cat14_sev3.mp4
+no20_cat23_sev4.mp4
+no21_cat13_sev4.mp4
+no22_cat23_sev4.mp4
+no23_cat18_sev3.mp4
+no24_cat13_sev4.mp4
+no25_cat23_sev4.mp4
+no26_cat32_sev4.mp4
+no27_cat21_sev3.mp4
+no28_cat23_sev4.mp4
+no29_cat21_sev3.mp4
+no2_cat18_sev3.mp4
+no2_cat23_sev4.mp4
+no3_cat22_sev4.mp4
+no3_cat23_sev4.mp4
+no4_cat32_sev4.mp4
+no5_cat19_sev4.mp4
+no5_cat5_sev2.mp4
+no6_cat5_sev2.mp4
+no7_cat23_sev4.mp4
+no7_cat29_sev3.mp4
+no8_cat21_sev3.mp4
+no8_cat22_sev4.mp4
+no9_cat14_sev3.mp4
+no9_cat22_sev4.mp4

videos.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5fe933e98bc7333bbddbb38a978803a4ebaef098e4629fae2ce7e2dcf04adcf7
+size 7117574683