captions/README.md

Japan Open Driving Dataset: Virtual Captioned Dataset

Overview

💡This repository provides a small subset of the Japan Open Driving Dataset.

The Japan Open Driving Dataset includes virtual captioned data for training vision-language (VLM) and vision-language-action (VLA) models for autonomous driving. This dataset covers the three key components of autonomous driving, referred to as 3P: Perception, Prediction, and Planning.

• Perception and prediction data are generated following the methodology of STRIDE-QA.
• Planning data is constructed following the approach of RACER.

The directory structure is organized as follows:

captions/
├── README.md
├── assets/
│   ├── ad08-2025-03-22-12-50-00__CAM_FRONT__1742615400407986.jpg
│   ├── ad08-2025-03-22-12-50-00__CAM_FRONT__1742615402007591.jpg
│   ├── ad08-2025-03-22-12-50-00__CAM_FRONT__1742615412006223.jpg
│   ├── ad08-2025-03-22-12-50-00__CAM_FRONT__1742615417005519.jpg
│   ├── ad08-2025-03-26-20-34-00__CAM_FRONT__1742988846968749.jpg
│   ├── ad20-2025-05-16-12-25-00__CAM_FRONT__1747365902002212.jpg
│   └── ad20-2025-05-16-12-26-00__CAM_FRONT__1747365961993178.jpg
├── STRIDE-QA/
│   ├── ego_centric_spatial_qa/
│   │   └── ego_centric_spatial_qa.jsonl
│   ├── ego_centric_spatiotemporal_qa/
│   │   └── ego_centric_spatiotemporal_qa.jsonl
│   └── object_centric_spatial_qa/
│       └── object_centric_spatial_qa.jsonl
└── RACER/
    └── racer.jsonl

STRIDE-QA

STRIDE-QA (Spatio-Temporal Reasoning In Driving Environments) is a visual question answering (VQA) dataset designed for physically grounded spatiotemporal reasoning in autonomous driving scenarios.

The dataset consists of three types of QA formats, as summarized in the table below:

Category	Description	# QAs
Object-centric Spatial QA	Spatial relations between two surrounding agents within a single frame. Includes both qualitative (e.g., relative position) and quantitative (e.g., distance, angle) queries.	3.45M
Ego-centric Spatial QA	Spatial relations between the ego vehicle and surrounding agents within a single frame. Covers distance, direction, and size comparisons.	8.20M
Ego-centric Spatiotemporal QA	Short-term prediction based on 4 context frames. Forecasts distance, heading angle, and velocity at t ∈ {1, 2, 3} seconds.	8.43M

Data Fields

The sample data included in this repository is provided in .jsonl. Each line represents a single sample with the following fields:

Field	Type	Description
id	str	Unique sample ID.
category	str	QA category type.
image	str	File name of the key frame used in the prompt.
images	list[str]	File names for the four consecutive image frames. Only available in Ego-centric Spatiotemporal QA category.
conversations	list[dict]	Dialogue in VILA format ("from": "human" / "gpt").
image_info	dict	Image metadata (height, width, dataset, keyframe_file_path).
token_info	dict	Sample and sample data tokens for nuScenes reference.
qa_info	list[dict]	Metadata for each QA turn (type, category, class, token).
bbox	list[list[float]]	Bounding boxes [x₁, y₁, x₂, y₂] for referenced regions.
rle	list[dict]	COCO-style run-length masks for regions.
region	list[list[int]]	Region tags mentioned in the prompt.

Example Records

The examples below are taken directly from the .jsonl files included in this directory.

The three STRIDE-QA examples below share the same key frame: samples/CAM_FRONT/ad08-2025-03-22-12-50-00__CAM_FRONT__1742615400407986.jpg.

Object-centric Spatial QA

{
  "image": "samples/CAM_FRONT/ad08-2025-03-22-12-50-00__CAM_FRONT__1742615400407986.jpg",
  "conversations": [
    {
      "from": "human",
      "value": "Between Region [2] and Region [1], which one appears on the right side from the viewer's perspective?"
    },
    {
      "from": "gpt",
      "value": "Region [2] is more to the right."
    },
    {
      "from": "human",
      "value": "How tall is Region [2]?"
    },
    {
      "from": "gpt",
      "value": "It is 1.84 meters."
    }
    // ... more QA pairs ...
  ],
  "image_info": {
    "height": 1240,
    "width": 1920,
    "dataset": "STRIDE-QA"
  },
  "token_info": {
    "sample_token": "3a5a7d1c7b0c380d480d414e40ef1b34",
    "sample_data_token": "46b247c98329e371a896c6f392fcb291"
  },
  "qa_info": [
    {
      "type": "qualitative",
      "category": "right_choice",
      "class": ["car", "car"],
      "token": {
        "obj_A": {
          "instance_token": "45cb739a26c7cf7f25bb463df9515d5a",
          "sample_annotation_token": "e0544ca6cb527addd3e6afea1d359206"
        },
        "obj_B": {
          "instance_token": "a5a3b6bd310cc09bae1ef7e1ab040066",
          "sample_annotation_token": "eed9dde32edaf19f27bc5da903a95101"
        }
      }
    }
    // ... more qa_info entries ...
  ],
  "bbox": [
    [703.7946166992188, 587.9367065429688, 1338.11328125, 1195.166015625],
    [431.61785888671875, 609.2484741210938, 787.3193969726562, 816.2525634765625],
    [1215.541748046875, 611.6207885742188, 1482.4794921875, 813.6328125]
  ],
  "rle": [
    {
      "size": [1240, 1920],
      "counts": "kc]k08WV1g1\\NQ3QM9E=\\Ob0F8I6M4K4K5J6K5K5H7L5G8K5J6M3N2L5J5M2M3L5M2O1N3L3N2O2M2O1N2N2N2O1O1O1N2O1O1O1O1N2O001O1O1O1O010..."
    }
    // ... more RLE masks ...
  ],
  "region": [
    [2, 1],
    [2, 1]
    // ... more region tags ...
  ]
}

Ego-centric Spatial QA

{
  "id": "46b247c98329e371a896c6f392fcb291",
  "category": "ego_centric_spatial_qa",
  "image": "samples/CAM_FRONT/ad08-2025-03-22-12-50-00__CAM_FRONT__1742615400407986.jpg",
  "conversations": [
    {
      "from": "human",
      "value": "Can you confirm if the ego vehicle is positioned to the left of Region [2]?"
    },
    {
      "from": "gpt",
      "value": "Indeed, the ego vehicle is positioned on the left side of Region [2]."
    },
    {
      "from": "human",
      "value": "Does the ego vehicle have a smaller size than Region [2]?"
    },
    {
      "from": "gpt",
      "value": "Actually, the ego vehicle might be larger than Region [2]."
    }
    // ... more QA pairs ...
  ],
  "image_info": {
    "height": 1240,
    "width": 1920,
    "dataset": "STRIDE-QA"
  },
  "token_info": {
    "sample_token": "3a5a7d1c7b0c380d480d414e40ef1b34",
    "sample_data_token": "46b247c98329e371a896c6f392fcb291"
  },
  "qa_info": [
    {
      "type": "qualitative",
      "category": "ego_left_predicate",
      "class": ["ego_vehicle", "car"],
      "token": {
        "ego": {
          "instance_token": null,
          "sample_annotation_token": null
        },
        "other": {
          "instance_token": "45cb739a26c7cf7f25bb463df9515d5a",
          "sample_annotation_token": "e0544ca6cb527addd3e6afea1d359206"
        }
      }
    }
    // ... more qa_info entries ...
  ],
  "bbox": [
    [703.7946166992188, 587.9367065429688, 1338.11328125, 1195.166015625],
    [431.61785888671875, 609.2484741210938, 787.3193969726562, 816.2525634765625],
    [1215.541748046875, 611.6207885742188, 1482.4794921875, 813.6328125]
  ],
  "rle": [
    {
      "size": [1240, 1920],
      "counts": "kc]k08WV1g1\\NQ3QM9E=\\Ob0F8I6M4K4K5J6K5K5H7L5G8K5J6M3N2L5J5M2M3L5M2O1N3L3N2O2M2O1N2N2N2O1O1O1N2O1O1O1O1N2O001O1O1O1O010..."
    }
    // ... more RLE masks ...
  ],
  "region": [
    [99, 2],
    [99, 2]
    // ... more region tags ...
  ]
}

Ego-centric Spatio-temporal QA

{
  "id": "46b247c98329e371a896c6f392fcb291",
  "category": "ego_centric_spatiotemporal_qa",
  "images": [
    "samples/CAM_FRONT/ad08-2025-03-22-12-50-00__CAM_FRONT__1742615400108087.jpg",
    "samples/CAM_FRONT/ad08-2025-03-22-12-50-00__CAM_FRONT__1742615400208053.jpg",
    "samples/CAM_FRONT/ad08-2025-03-22-12-50-00__CAM_FRONT__1742615400308020.jpg",
    "samples/CAM_FRONT/ad08-2025-03-22-12-50-00__CAM_FRONT__1742615400407986.jpg"
  ],
  "conversations": [
    {
      "from": "human",
      "value": "How distant is Region [0] from the ego vehicle?"
    },
    {
      "from": "gpt",
      "value": "The ego vehicle and Region [0] are 3.09 meters apart."
    },
    {
      "from": "human",
      "value": "Provide the ego vehicle's current speed."
    },
    {
      "from": "gpt",
      "value": "The speed of the ego vehicle is 0.38 m/s."
    }
    // ... more QA pairs ...
  ],
  "image_info": {
    "height": 1240,
    "width": 1920,
    "dataset": "STRIDE-QA",
    "keyframe_file_path": "samples/CAM_FRONT/ad08-2025-03-22-12-50-00__CAM_FRONT__1742615400407986.jpg"
  },
  "token_info": {
    "current": {
      "sample_token": "3a5a7d1c7b0c380d480d414e40ef1b34",
      "sample_data_token": "46b247c98329e371a896c6f392fcb291"
    },
    "prev_1": {
      "sample_token": "c56137d0c67c28deab0d5dee2f9b5ad9",
      "sample_data_token": "4d8e2f82573c8cdf7b28f82d0ff6f1c8"
    },
    "prev_2": {
      "sample_token": "cff78b51d578e826de9df11b83bd581a",
      "sample_data_token": "0d895838976397670e026bb0e577ec35"
    },
    "prev_3": {
      "sample_token": "2d6d8970fa9ac3a93849d1e2b484a9b6",
      "sample_data_token": "e50b18039fe3de5c3d1d92a1d65df2a5"
    }
  },
  "qa_info": [
    {
      "category": "target_distance_t0",
      "target": {
        "instance_token": "6642580b2804a97393f62405997f3f5a",
        "sample_annotation_token": "d083e83c91e49e05d99c29e06b9a65ce",
        "class": "car",
        "region_id": 0
      },
      "state": {
        "ego": {
          "speed_mps": 0.3785464964835892
        },
        "target": {
          "speed_mps": 0.16133338718362417,
          "distance_m": 3.093587157841829,
          "bearing_deg": -3.1617669265292534,
          "clock_position": 12
        },
        "after_t_seconds": 0
      }
    }
    // ... more qa_info entries ...
  ],
  "bbox": {
    "current": [
      [703.7946166992188, 587.9367065429688, 1338.11328125, 1195.166015625],
      [431.61785888671875, 609.2484741210938, 787.3193969726562, 816.2525634765625],
      [1215.541748046875, 611.6207885742188, 1482.4794921875, 813.6328125]
    ],
    "prev_1": [
      [717.9148083999381, 585.4938226749672, 1328.141118240055, 1190.9798084989009],
      [441.79452219562694, 609.7145600574536, 788.4789893908609, 818.3366613264577],
      [1219.0347358857923, 610.7476111116105, 1484.2271835194078, 810.5061656447125]
    ]
    // ... prev_2, prev_3 ...
  },
  "rle": {
    "current": [
      {
        "size": [1240, 1920],
        "counts": "kc]k08WV1g1\\NQ3QM9E=\\Ob0F8I6M4K4K5J6K5K5H7L5G8K5J6M3N2L5J5M2M3L5M2O1N3L3N2O2M2O1N2N2N2O1O1O1N2O1O1O1O1N2O001O1O1O1O010..."
      }
      // ... more current-frame masks ...
    ],
    "prev_1": [
      {
        "size": [1240, 1920],
        "counts": "oZXj147McU1Q1H6H8D;C]N`kNg1ZT1>N3M3N2M3L5K4L5M5K6J5J4J6K5K4M3M3M3K5M3M2N2O1N3O000O2O00000O10000O10000000O1000000..."
      }
      // ... more previous-frame masks ...
    ]
    // ... prev_2, prev_3 ...
  }
}

RACER

RACER (Rationale-Aware Captioning of Edge-Case Driving Scenarios) leverages a multi-stage, VLM-based automatic annotation pipeline to generate textual rationales describing the ego vehicle’s driving behavior. It contains a total of 81,204 samples.

Data Fields

The sample data included in this repository is provided in .jsonl. Each line represents a single record with the following fields:

Field	Type	Description
id	str	Unique sample data token.
image	str	Front-camera image path associated with the record.
token_info	dict	nuScenes-compatible scene, sample, and sample data tokens.
reasoning_description	str	Natural-language rationale describing the ego vehicle's behavior.
trajectory	list[list[float]]	Planned ego trajectory represented as a sequence of 15-dimensional state vectors.

The sample records below are taken directly from captions/RACER/racer.jsonl.

Sample 1

{
  "id": "247c6664cfe3c995e016fe278cefa1f5",
  "image": "samples/CAM_FRONT/ad20-2025-05-16-12-25-00__CAM_FRONT__1747365902002212.jpg",
  "token_info": {
    "scene_token": "6da5a4524c3757c34740168d8381cbbc",
    "sample_token": "071925c0c29c0b809949a4abf31520b5",
    "sample_data_token": "247c6664cfe3c995e016fe278cefa1f5"
  },
  "reasoning_description": "Yield to pedestrians actively crossing the road in the crosswalk.",
  "trajectory": [
    [0, 0, 0, 1.8018463850021362, -0.0014932957710698247, -0.02612774632871151, 0.6270471215248108, 0.020551223307847977, -9.857547760009766, -3.875173665736893e-18, 2.5432877684588695e-18, -3.946085295391165e-17, -0.0026144690345972776, 0.0065924013033509254, 0.009308919310569763],
    [0.017622489482164383, -2.032950214925222e-05, -0.0002920156402979046, 1.8116471767425537, -0.002634715288877487, -0.026769571006298065, 0.6710872650146484, -0.016269853338599205, -9.876188278198242, -3.1003208277979866e-05, 7.769500371068716e-05, 9.448604396311566e-05, -0.0031095994636416435, 0.007979712449014187, 0.009636128321290016],
    // ...
    [12.19664478302002, -23.181495666503906, -0.32858118414878845, 5.638118267059326, -0.11004285514354706, -0.09804873913526535, 0.21572671830654144, 0.5433729887008667, -9.714394569396973, -0.05320020020008087, -0.009889491833746433, -1.6617497205734253, -0.01800353080034256, 0.022977812215685844, 0.03307747840881348]
  ]
}

Sample 2

{
  "id": "b0b42e07d1e5684b20775d782815a056",
  "image": "samples/CAM_FRONT/ad20-2025-05-16-12-26-00__CAM_FRONT__1747365961993178.jpg",
  "token_info": {
    "scene_token": "11c0f67419cd607a8a90850cf003295a",
    "sample_token": "138b544316f0169629c34469426a073d",
    "sample_data_token": "b0b42e07d1e5684b20775d782815a056"
  },
  "reasoning_description": "Wait until the oncoming vehicle passes before proceeding, as there is a parked vehicle ahead.",
  "trajectory": [
    [-2.5849394142282115e-26, -1.0501316370302109e-26, 0, 0.0007999999797903001, 0.0008999999845400453, -0.000699999975040555, -0.18780891597270966, 0.18488502502441406, -9.799281120300293, -2.5388181497177275e-19, -1.575437027730227e-18, -4.868243778494266e-17, -1.70063412952004e-05, -0.0015078369760885835, 0.00046040245797485113],
    [-2.5849394142282115e-26, -1.0501316370302109e-26, 0, 0.0007999999797903001, 0.0008999999845400453, -0.000699999975040555, -0.22928227484226227, 0.18567614257335663, -9.810667991638184, -2.5388181497177275e-19, -1.575437027730227e-18, -4.868243778494266e-17, 0.0002695784787647426, -0.0026303157210350037, 8.236600842792541e-05],
    // ...
    [4.0264387130737305, 0.6593005657196045, -0.11963625997304916, 3.884610414505005, 0.11566238850355148, -0.0688113197684288, 1.8854002952575684, 0.15707406401634216, -9.889180183410645, 0.0035211581271141768, 0.005016455892473459, 0.19437487423419952, 0.0034317863173782825, -0.00033042821451090276, 0.10067762434482574]
  ]
}

Sample 3

{
  "id": "49fe76eaff0e03ae614e85dc11d6668e",
  "image": "samples/CAM_FRONT/ad08-2025-03-26-20-34-00__CAM_FRONT__1742988846968749.jpg",
  "token_info": {
    "scene_token": "d46b258e61c58fefe7764a915603ac17",
    "sample_token": "996c7d673e5bf20acf9f5336edb614fb",
    "sample_data_token": "49fe76eaff0e03ae614e85dc11d6668e"
  },
  "reasoning_description": "Proceed straight while exercising caution, as the bicycle ahead on the left is protruding into the roadway.",
  "trajectory": [
    [2.5849394142282115e-26, 5.169878828456423e-26, 0, 4.0771002769470215, 0.00015322334365919232, -0.06909190118312836, 1.582289695739746, 0.05637861043214798, -9.653964042663574, 1.4352722035113056e-18, 3.0090239541410324e-18, 8.444341442930057e-17, 0.0031817536801099777, -0.00031303070136345923, -0.03233116865158081],
    [0.03989756852388382, 2.139501884812489e-05, -0.0006439747521653771, 4.0937089920043945, -0.0005220293533056974, -0.06808898597955704, 1.4368404150009155, 0.058929115533828735, -9.701728820800781, 2.6970807084580883e-05, -3.8133155612740666e-05, -0.000330952403601259, 0.002704455517232418, -0.004207425285130739, -0.033908676356077194],
    // ...
    [52.65324401855469, -2.3958933353424072, -0.8994203805923462, 3.638854742050171, -0.11903553456068039, -0.06169835105538368, -0.7993940711021423, -0.41783344745635986, -9.790143966674805, 0.015364415012300014, 0.0021477406844496727, -0.08191817253828049, -0.0019924442749470472, 0.007094692904502153, -0.09190614521503448]
  ]
}