The project is about using the Multiverse model to make possible multiple-future trajectory prediction for a seen person and making it ready for deploying in a slow self-driving car by adding the required preprocessing like :

• annotation and object tracking (by using YOLOv5 and DeepSORT)
• semantic segmentation (by using DeepLab)

and making the whole process automated.

the pipeline in local machine for testing and evaluation :

Inferencing on a real video.

final model output on a test video :