Question regarding the value of target_point compared to the values of the predicted points of the model: Target Point and Predicted Waypoints Comparison

[MCUBE-2023]

Dear author,

My objective is to compare the pred_wp (returned by the function forward_ego) with the actual values of the waypoints (ground truth values of waypoints). To this end, I have the following questions:

1- Which variable contains the actual values of the waypoints (ground truth values of waypoints) ?
2- When I printed the size of pred_wp in the submission agent file, it showed tensor torch.Size([1, 4, 2]). What does these values (1,4,2) mean ?
3- If we want to compare pred_wp to its equivalent ground truth, what is the equivalent ground truth of pred_wp ?

[Kait0]

2 - batch size 1, 4 waypoints predicted, each waypoint consits of 2 coordinates x and y.
You can only really compare to ground truth if you are working with a static dataset (where you know the future).
During inference you don't have ground truth waypoints available.
You could make them available maybe if you run the autopilot as well simulate it for 8 seconds, then roll back the simulation and compare with that. But that is super complicated and I don't think you can perfectly reset CARLA in version 0.9.10.