{
  "fetched_at": "2026-10-20T06:57:20Z",
  "source_url": "https://www.kaggle.com/competitions/camera-submission/data",
  "pages": [
    {
      "name": "data-description",
      "content": "The vehicle–infrastructure–air (VIA) collaborative perception dataset, collected through the simulation platform, contains multi-view images and point cloud data from connected autonomous vehicles, roadside infrastructure, and aerial platforms. The dataset provides annotated ground-truth bounding boxes, sensor calibration parameters, and temporal sequence information.\nParticipants are required to write the inference results for each frame in the test set into a CSV file (see example below) following the specified format. The submitted CSV will be compared with the ground-truth (GT) data in the system to compute the final score.\nSubmission Format Instructions\nThe CSV file must contain only two columns:\n\nFirst column: ID—the sequential index of each test frame (a total of 2170 frames).\n\n\nSecond column: all vehicle detection results for that frame (one or more targets).\n\nFor each frame, the predicted bounding boxes from the object detection model should be listed in order within the second column.\nEach predicted bounding box must include the following values in the specified order:\nvehicle score center_x center_y center_z length, width, height, rotation_x, rotation_y, rotation_z\n\nwhere each item represents:\n\nvehicle—object category\nscore—confidence score\ncenter_x, center_y, center_z — 3D bounding box center coordinates\nlength, width, height—bounding box dimensions\nrotation_x, rotation_y, rotation_z—roll, yaw, and pitch angles\n\n⚠️ Important:\nThe CSV file must contain exactly two columns.\nIf there are multiple targets in a frame, all should be listed sequentially within the same cell of the second column.\nAny formatting errors in the CSV file will result in failure to obtain a score.\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "Competition Rules\n\n--Competition Schedule--\n\nDeadline for submission: October 25, 2025\n\n--Task Description--\n\nThis challenge focuses on key technologies such as multi-source sensing data processing, information fusion, and collaborative perception across vehicles, infrastructure, and aerial platforms. The main goals are to address issues of spatio-temporal alignment and calibration of multi-agent perception data, precise fusion of heterogeneous sensors, and cross-view spatial relationship modelling of detected targets.\n\nParticipants are required to consider the complex and highly dynamic nature of the physical world, as well as the heterogeneity of multi-agent data. Using the provided vehicle-infrastructure-air (VIA) collaborative perception dataset, teams should design models capable of adapting to various conditions (e.g., lighting, weather, viewing angles). They must train and validate their models using the supplied aerial–ground cooperative perception dataset and submit the testing code and documentation accordingly.\n\nThe expected submissions should demonstrate algorithmic innovation, high perception accuracy, and robust cross-view alignment, thereby advancing research and applications in the field of vehicle–infrastructure–aerial collaborative perception.\n\n--Problem Modelling--\n\nInput: Vehicle-mounted LiDAR point clouds, roadside LiDAR point clouds, and scene target annotation files.\nOutput: Target categories and poses within the scene.\n\n--Other Rules--\n\n(1) Originality: Entries must be original and comply with all relevant laws of the People’s Republic of China. They must not infringe on any third-party intellectual property or other rights. Once plagiarism or rights violations are verified, the organizing committee will disqualify the team and revoke the results. The same work may not be submitted for multiple tasks—any repetition will be deemed invalid.\n\n(2) Fair Competition: Participants are prohibited from exploiting rules or technical loopholes to gain unfair advantages, from copying others’ work, exchanging answers, or using multiple accounts. Any such misconduct will result in disqualification and disciplinary action.\n\n(3) Organizer’s Statement: The organizing committee reserves the right to modify the competition rules, determine and handle cases of cheating, and withdraw or withhold awards from teams whose behavior compromises fairness or organizational integrity.\n\n(4)  Intellectual Property: The intellectual property of all submitted works (including but not limited to algorithms, models, and solutions) shall be jointly owned by the participants and the organizing committee. The committee reserves the right to use the submissions, competition information, and team details in promotional materials, publications, authorized media releases, official website displays and downloads, exhibitions, or roadshows. The committee also retains priority rights for investment facilitation and product incubation based on submitted works.\n\n(5)  Platform Security: Any attacks on the competition platform will result in immediate disqualification and forfeiture of awards. The committee reserves the right to pursue legal responsibility for such actions.\n\n(6)  Special Rules: Participants must submit two algorithms separately—one based on camera data and one based on LiDAR point cloud data—via the designated submission links. Any use of mixed or additional modalities at the input stage will result in disqualification once detected. The final score is calculated by weighting the camera method score and the LiDAR method score at a ratio of 60%:40%.\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "For the evaluation of camera-based models, the Average Precision (AP) metric for object detection is used.\nThe detection accuracy is calculated under two different Intersection over Union (IoU) thresholds: 0.5 and 0.7.\nThe final AP score is then obtained as a weighted combination of the two results, as follows:\n\nAP=40%*AP@50+60%*AP@70。",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "For specific rules, please comply with the Confidentiality Agreement for the Intelligent Transportation Association Innovation Challenge.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "abstract",
      "content": "The aim of this Vehicle-Infrastructure-Air Cross-Domain Collaborative Perception Algorithm Competition is to significantly enhance the environmental perception capabilities of intelligent unmanned systems in intricate, highly dynamic, and unstructured road scenarios.  Participants in this challenge must create clever algorithms for effective cross-domain data fusion and collaborative perception.  Multi-source, heterogeneous data from connected-autonomous vehicles, mobile roadside infrastructure units, and aerial platforms must be integrated into these algorithms.  The objective is to accomplish cross-domain data fusion and collaborative perception in complicated settings by overcoming the constraints of single-platform perception.  In the end, this will give autonomous unmanned systems stronger assistance for job planning and wise decision-making by improving their environmental sensing capabilities.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}