Why is my trajectory error so large?

[Alex-fishred]

thank you for a great job!!!
I have tried throwing past trajectories into the model, but the predicted trajectories are not ideal.
As shown in the figure below, I make trajectory predictions for motorcycles, cars, and bicycles

What is the input/output of the model?
I have observed that the model needs to input 38 data
How are these 38 pens composed?
The center point of the target (21 list) or the target box (41 list)?
What are the 4 output data? Target center point? Target box?

[TomatoFT]

Hi @Alex-fishred, thank you for your interested in my work.
I will answer your questions.
What is the input/output of the model?

• The input of the model is the type of vehicle (One hot encoder for 5 types of the vehicle (Motorbike, Bicycle, Car, Truck and Bus) ) and the coordinate of 7 period of frames of the vehicle.

features = ['Bicycle', 'Bus', 'Car', 'Motorcycle', 'Truck',
            'X_min', 'Y_min', 'X_max', 'Y_max',
            'X_min_1', 'Y_min_1', 'X_max_1', 'Y_max_1',
            'X_min_2', 'Y_min_2', 'X_max_2', 'Y_max_2',
            'X_min_3', 'Y_min_3', 'X_max_3', 'Y_max_3',
            'X_min_4', 'Y_min_4', 'X_max_4', 'Y_max_4',
            'X_min_5', 'Y_min_5', 'X_max_5', 'Y_max_5',
            'X_min_6', 'Y_min_6', 'X_max_6', 'Y_max_6',
            'X_min_7', 'Y_min_7', 'X_max_7', 'Y_max_7',
            ]

target = ['X_max_8', 'X_min_8', 'Y_max_8', 'Y_min_8']

And the output data is 4 coordination of the next periods.

the problems about why the trajectory error is so large.

• Actually the trajectory model is not be retrained in the large dataset. It's only contain 178881 observation (the trajectory collect from 3 1-hour-length videos).
• Moreover the features is not diverse also because it's only contain the trajectory and the type of the vehicle. There is more things that need to have like the type of obstacles, the coordination of the lane, the traffic light, etc. So this is just the experiment, if there is more feature the model can learn better.

If you have anymore concerns, feel free to let me know.