To increase the size of the AI ​​dataset, I decided to add a video. To do this, I retrieved a video from YouTube of one of the races of the 2023 season. This video was provided in the roboflow API. Using this API, I then manually annotated each frame of the video. In this way the equivalent of 400 images were added to the database.

Example of annotation:

Result :

The AI ​​is able to recognize more cars at the same time, with more precise car delineation. Additionally, my AI now supports videos. Thanks to video, the AI ​​no longer detects the leaderboard like an F1 and detection artifacts outside the circuit were greatly reduced. It would be possible to increase the accuracy of the model even further in the future.

Version française

Ajout d’une vidéo dans la database d’une IA