FAQ
A: The videos of the games provided in SoccerNet contains copyright. The NDA is a contract between the users and the providers to prevent the re-distribution of copyright materials.
A: The videos of the games provided in SoccerNet contains copyright. The NDA is a contract between the users and the providers to prevent the re-distribution of copyright materials.
Q: Can I train my own model with the SoccerNet dataset?
Q: Can I train my own model with the SoccerNet dataset?
A: Of course! We invite you to train models on the SoccerNet dataset, and if you end up publishing a paper, please cite our work!
A: Of course! We invite you to train models on the SoccerNet dataset, and if you end up publishing a paper, please cite our work!
Q: Is it fine to include screenshots/clips of the soccer videos in research papers/presentations?
Q: Is it fine to include screenshots/clips of the soccer videos in research papers/presentations?
A: Yes, as long as you are refering to the source of your data, you can include screenshots/clips of the videos in research papers/presentations.
A: Yes, as long as you are refering to the source of your data, you can include screenshots/clips of the videos in research papers/presentations.
Q: Can I use the data from SoccerNet for commercial purposes?
Q: Can I use the data from SoccerNet for commercial purposes?
A: No. The SoccerNet dataset is meant for research purposes, it is not intended for commercial purposes. The videos of SoccerNet contains copyright that belongs to each European leagues, if you want to a create a business around soccer video understand, you will have to collect your own videos. The SoccerNet team cannot facilitate this process for you.
A: No. The SoccerNet dataset is meant for research purposes, it is not intended for commercial purposes. The videos of SoccerNet contains copyright that belongs to each European leagues, if you want to a create a business around soccer video understand, you will have to collect your own videos. The SoccerNet team cannot facilitate this process for you.
Q: Can I use the algorithms developed by the SoccerNet team for commercial purposes?
Q: Can I use the algorithms developed by the SoccerNet team for commercial purposes?
A: Each algorithm has its own licence, please visit the github repository and read the license to understand the limitation of the usage for each algorithms.
A: Each algorithm has its own licence, please visit the github repository and read the license to understand the limitation of the usage for each algorithms.
Q: Can you clarify the data usage during the challenge period?
Q: Can you clarify the data usage during the challenge period?
A:
1. On the restriction of private datasets and additional annotations
A:
1. On the restriction of private datasets and additional annotations
SoccerNet is designed to be a research-focused benchmark, where the primary goal is to compare algorithms on equal footing. This ensures that the focus remains on algorithmic innovation rather than data collection or annotation effort. Therefore:
SoccerNet is designed to be a research-focused benchmark, where the primary goal is to compare algorithms on equal footing. This ensures that the focus remains on algorithmic innovation rather than data collection or annotation effort. Therefore:
- Any data used for training or evaluation must be publicly accessible to everyone to prevent unfair advantages.
- By prohibiting additional manual annotations (even on publicly available data), we aim to avoid creating disparities based on resources (e.g., time, budget, or manpower). This aligns with our commitment to open-source research and reproducibility.
2. On cleaning or correcting existing data
2. On cleaning or correcting existing data
We recognize that publicly available datasets, including SoccerNet datasets, might have imperfections in their labels (around 5% usually). Cleaning or correcting these labels is allowed outside of the challenge period to ensure fairness:
We recognize that publicly available datasets, including SoccerNet datasets, might have imperfections in their labels (around 5% usually). Cleaning or correcting these labels is allowed outside of the challenge period to ensure fairness:
- Participants can propose corrections or improvements to older labels before the challenge officially starts. Such changes will be reviewed and potentially integrated into future versions of SoccerNet. Label corrections can be submitted before or after the challenge for inclusion in future SoccerNet releases, ensuring a fair and consistent dataset during the competition.
- During the challenge, participants should not manually alter or annotate existing labels, as this introduces inconsistency and undermines the benchmark's fairness.
- Fully automated methods for label refinement or augmentation, however, are encouraged. These methods should be described in the technical report to ensure transparency and reproducibility.
3. Defining “private datasets”
3. Defining “private datasets”
A dataset is considered “private” if it is not publicly accessible to all participants under the same conditions. For example:
A dataset is considered “private” if it is not publicly accessible to all participants under the same conditions. For example:
- Older SoccerNet data are not private, as they are available to everyone.
- However, manually modifying or adding annotations (e.g., bounding boxes or corrected labels) to older SoccerNet data during the challenge creates a disparity and would be considered "private" unless those modifications are shared with the community in advance.
4. Creative use of public data
4. Creative use of public data
We fully support leveraging older publicly available SoccerNet data in creative and automated ways, as long as:
We fully support leveraging older publicly available SoccerNet data in creative and automated ways, as long as:
- The process does not involve manual annotations.
- The methodology is clearly described and reproducible.
- For instance, if you develop an algorithm that derives additional features or labels (e.g., bounding boxes) from existing data, this aligns with the challenge's goals and is permitted.
5. Data sharing timeline:
5. Data sharing timeline:
To ensure fairness, we decided that any new data must be published or shared with all participants through Discord at least one month before the challenge deadline. This aligns with the CVsports workshop timeline and allows all teams to retrain their methods on equal footing.
To ensure fairness, we decided that any new data must be published or shared with all participants through Discord at least one month before the challenge deadline. This aligns with the CVsports workshop timeline and allows all teams to retrain their methods on equal footing.