The traditional approach (top) relies on object detection, depth estimation, and optical flow processed through a Graph Convolutional Network for accident anticipation. In contrast, our proposed ...