SPECIFIC CHALLENGE TASKS
Fig. 2. We propose a Rider Intention Prediction (RIP) challenge aimed at enhancing the safety of two-wheeler riders. We introduce two tasks in our challenge: single-view RIP (Task 1) and multi-view RIP (Task 2), where we utilize our proposed dataset to benchmark the performance.
Task 1: Single-view rider intention prediction
For training, we offer a set of 500 exterior single-view (frontal-view) videos. During the validation phase, an additional 200 videos will be released as the validation set. As the competition progresses, participants will receive 300 test videos before its conclusion. Evaluation will be conducted using two widely accepted metrics, namely F1 and Accuracy, to assess predictive performance.
Task 2: Multi-view rider intention prediction
The dataset includes 500 training videos with a multi-view exterior traffic context, encompassing frontal-view, left side-mirror view, and right side-mirror view. During the validation phase, an additional 200 videos are released as validation set, and 300 test videos will be released before the competition concludes. Evaluation of predictions will be based on two commonly used metrics: F1 and Accuracy.
We encourage teams to participate in both tasks, although they may choose to compete in either one or both of them.