{
  "fetched_at": "2026-10-23T03:31:27Z",
  "source_url": "https://www.kaggle.com/competitions/ava-challenge/data",
  "pages": [
    {
      "name": "data-description",
      "content": "Your challenge in this competition is to forecast the car accident. In this dataset, we provide the videos of roads and freeways to let you train the model.\n\nPlease use training dataset to create your own model, and make classification on testing dataset. The name of each film is generated randomly. For your convenience, we named the road data after 20001 to 20157, while freeway is 30001 to 30155. Each film has a jpg folder and a xml folder to show annotations.\n\nRemember to fill out the request form that we will send you the data link. \n\nRequest Form: https://docs.google.com/forms/d/1f30zkr9iVMoYbLq9sFD8t5LCmMdd_7NT7Hgkr8pJJLE/edit\n\n## Files\n\n*   **TRAIN DATASET** - the training set\n         |-------  **freeway train** - the training set of freeway\n         |            |-------  **jpgs** - the videos\n         |            |-------  **xmls** - the annotations\n         |-------  **road train** - the training set of road\n         |            |-------  **jpgs** - the videos\n         |            |-------  **xmls** - the annotations\n         |-------  **freeway train.csv** - labels of videos in the freeway folder\n         |-------  **road train.csv** - labels of videos in the road folder\n*   **TEST DATASET** - the testing set\n         |-------  **freeway test** - the testing set of freeway\n         |            |-------  **jpgs** - the videos\n         |            |-------  **xmls** - the annotations\n         |-------  **road test** - the testing set of road\n         |            |-------  **jpgs** - the videos\n         |            |-------  **xmls** - the annotations\n*   **sample submission.csv** - the example of submission format\n\n## Columns\n\n*   `film name` - name of video (randomly generated)\n*   `car accident` - the anwsers, show there will be a car accident or not. 1 indicates yes, while 0 indicates no.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "## Announcement of Challenge Winners\n\nWe're excited to reveal the top performers in the Anticipating Vehicle Accidents Challenge! The following participants will be awarded certificates, and the first-place winner will receive an additional prize. Congratulations!!!\n\n---\n\n1st Place: Matthew1012, with a score of 0.67142\nSolution: Predicting Car Accidents with YOLOv7 Object Detection and Object Relationships\n2nd Place: ALLAccept, with a score of 0.61428\nSolution: Improved Dynamic Spatial-Temporal Attention Network for Early Anticipation of Traffic Accidents\n\n---\n\n## Private Leaderboard Score Released\nWe are pleased to announce the top three performers on our private leaderboard for the recent competition. These participants have achieved remarkable scores.\n\n---\n\n1st Place: Matthew1012, 0.67142\n2nd Place: Huang Yu Hung, 0.61904\n3rd Place: ALLAccept, 0.61428\n4th Place: asd3200asd, 0.59047\n5th Place: Host(not included)\n6th Place: abida gurman, 0.58571\n\n---\n\nWe will send an email to the above participants to notify them to submit a link to their technical report (1 page) and source code. Please stay tuned for this important message.\nAll of the aforementioned participants are required to submit their challenge papers. Please note the deadline for paper submission and we encourage all participants to submit their work as well.\n\n**4/21 Latest News!!**\n1. The deadline for **Final Submission** is based on UTC time (8:00 a.m. UTC+8). Please ensure to upload the files in time, thank you!\n2. The paper submission deadline has been moved up from **May 1** to **April 28** due to ICME committee decision. We will send out the acceptance notice of your papers on April 29. Participants are kindly reminded to take note of this change.\n3. Camera-ready deadline is May 1 UTC time.\n\n## Goal of the Competition\nVideo analysis research is crucial for autonomous driving as it can assist with accident prediction. Dashcam videos can offer valuable information for this task. However, there is a scarcity of well-annotated, large-scale datasets for accident prediction in Asian countries.\nWe collected and labeled a large-scale vehicle collision dataset from Taiwan to address this gap for frame- and video-level predictions. The dataset aims to inspire participants from academia and industry to enhance safety in autonomous driving. The track of this challenge is vehicle accident prediction. Please try to predict whether there will be a car accident within the next 20 frames. More information can be found on the website [https://sites.google.com/view/tvcd-tw/ava-challenge?authuser=0](https://sites.google.com/view/tvcd-tw/ava-challenge?authuser=0)\n**In order for award recipients to participate in the IEEE ICME2023 Grand Challenge AVA Special Session and present their work, they are required to submit challenge papers and give their consent. Failure to submit the paper or submitting a paper that is less than 3 pages in length will result in the cancellation of the award.**\n\n\n\n\n\n\n## Context\nAccording to the research, self-driving cars can only reduce about one-third of major car accidents that occur in the United States out of 5,000 incidents. People typically assume that autonomous driving systems can eliminate human errors while driving. However, many car accidents are still caused by human factors, including perception errors, driver distraction, poor visibility, or slow reactions. Other possible causes are prediction errors, such as misjudging the distance to other vehicles or pedestrians' movements. Some errors may be related to planning and decision-making, such as inappropriate evasive actions or overcompensation when controlling the vehicle. Other accidents can be caused by impairment due to alcohol, drugs, or drowsiness. There are avoidable accidents that are caused by vehicle malfunctions.\nBased on the above, we propose a topic for participants to use existing data to predict whether a car accident will occur in the next second to facilitate crisis response in self-driving and improve traffic safety.\nThe competition host is Advanced Computer Vision Lab (ACVLab), a research team led by Professor Chih-Chung Hsu at National Cheng Kung University. Our lab focuses on introducing machine learning/deep learning into various applications in computer vision and cultivating students with high practical and applied capabilities in deep learning.\nYour efforts in this competition could help extend the benefits of road safety. Greater access could further reduce the occurrence of car accidents worldwide.\n\n## Contact Information\nIf you have any questions, please contact the following ACVLab members.\nYun-Zhong Jiang (yunzhong1105@gmail.com)\nWei-Hao Huang (robert20000831@gmail.com)\n\n## Rank\nThe valid submission only if the participants have submitted their report, code, and successful submission on kaggle could be considered as the winner. For the winners of the competition (top three), we will invite the top three to five to share their winning solutions at ICME and award certificates to the top three. For other information could refer to our official website https://sites.google.com/view/tvcd-tw/ava-challenge?authuser=0\n\n\n\n\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "## Goal\nIt is your job to predict there's a car accident or not. For each film name in the test set, you must predict the outcome of the `car accident` variable. \n\n## Metric\nThe evaluation metric for this competition is [**Mean F1-Score**](https://en.wikipedia.org/wiki/F-score). The F1 score, commonly used in information retrieval, measures accuracy using the statistics precision **p** and recall **r**.\n\nPrecision is the ratio of true positives **tp** to all predicted positives **tp** **+** **fp**. Recall is the ratio of true positives **tp** to all actual positives **tp** **+** **fn**. The **F1-score** is given by:\n\n$$ \\text{F1} = 2\\frac{\\text{p} \\cdot \\text{r}}{\\text{p}+\\text{r}}\\ \\ \\mathrm{where}\\ \\ \\text{p} = \\frac{\\text{tp}}{\\text{tp}+\\text{fp}},\\ \\ \\text{r} = \\frac{\\text{tp}}{\\text{tp}+\\text{fn}} $$\n\nThe F1 metric weights recall and precision equally, and a good retrieval algorithm will maximize both precision and recall simultaneously. Thus, moderately good performance on both will be favored over extremely good performance on one and poor performance on the other.\n\n## Submission Format\n\nPlease submit the file in .csv format that includes two features: `film name` and `car accident`. You may refer to the dataset for an example submission file.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "These are the rules of this competition. Participants must agree to the rules before joining. The rules are as follows:\n\n - To promote the development of novel and relatively simple models for accident prediction, model ensembles are NOT allowed in this competition.\n - Team mergers are allowed and the maximum team size is 6.\n - You cannot create or use multiple Kaggle accounts to submit your solutions.\n - You cannot privately share code or data outside of your team. You can share code publicly on the forums if you wish.\n - Please fill out the request form and we will send you the data link. [https://docs.google.com/forms/d/1f30zkr9iVMoYbLq9sFD8t5LCmMdd_7NT7Hgkr8pJJLE/edit](url)\n - Don't cheat!\n ",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Timeline",
      "content": "**Latest News!!**\n1. The deadline for **Final Submission** is based on UTC time (8:00 a.m. UTC+8). Please ensure to upload the files in time, thank you!\n2. The paper submission deadline has been moved up from **May 1** to **April 28** due to ICME committee decision. We will send out the acceptance notice of your papers on April 29. Participants are kindly reminded to take note of this change.\n3. Camera-ready deadline is May 1 UTC time.\n\n**Final extension!!**\n\nThe latest news will update on our website synchronously. Don't forget to check regularly. https://sites.google.com/view/tvcd-tw/ava-challenge\n\nIn order for award recipients to participate in the IEEE ICME2023 Grand Challenge AVA Special Session and present their work, they are required to submit challenge papers and give their consent. Failure to submit the paper or submitting a paper that is less than 3 pages in length will result in the cancellation of the award.\n\n\n\n* **March 3, 2023** - Start Date. \n~~* **April 10, 2023** - Final Submission Deadline~~.\n* **April 21, 2023** - Final Submission Deadline.\n~~* **April 11, 2023** - Private Leader Board Announce.~~\n* **April 22, 2023** - Private Leader Board Announce.\n~~* **April 17, 2023** - Technical Report Submission Deadline~~.\n* **April 24, 2023** - Technical Report Submission Deadline\n~~* **April 22, 2023** - Competition End Date - Winner's announcement~~.\n* **April 25, 2023** - Competition End Date - Winner's announcement\n~~* **May 1, 2023** - Papers Submission Deadline (All teams are invited)~~\n* **April 28, 2023** - Papers Submission Deadline (All teams are invited)\nPaper submission: https://cmt3.research.microsoft.com/User/Login?ReturnUrl=%2FICMEW2023\n* **May 1, 2023** - Camera-ready deadline\n\nAll deadlines are at 11:59 PM UTC+8 on a corresponding day unless otherwise noted. The competition organizers reserve the right to update the contest timeline if they deem it necessary.\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}