issue about only one agent prediction

[jhtao1860]

@xincoder thanks for sharing your algorithm. I have a question, for only one-agent, the prediction looks no good. Do you know the reason?

[xincoder]

Hi @jhtao1860 , thank you for your interest in our work and run it on your side. I did not face this problem before. There are several things that we may want to double-check: (1) While you are predicting the trajectories, please make sure the data of valid traffic objects are put in the first raws. (2) Data scale. The observed history (blue) and predicted trajectory (red) should have the same scale. If there is no problem with the above-mentioned parts, you should not get such kind of outputs.

[jhtao1860]

@xincoder the result is the 12th( index 0 is the first ) case of all 415 scenarios in "prediction_test.txt" . I did not change the source code so I did not change the "Data scale"（and dont know how to change it）. I found some phenomenon. ONLY the predicition of one-agent-in-a-scenario is not good. SO I tried 2 ways to cope the problem. 1. Make a static-fake agent to get 2 agent in the case. 2 Change the in_channels to 2. Both the ways get good prediction(though i dont know why)

[xincoder]

@jhtao1860 Thank you for clarifying. I did not visualize all testing samples before, thus I thought you run the code on your own data. Sorry about the misunderstanding. In this case, it may be caused by the trained-weight. Maybe a single object causes a numerical stability problem.

[lylyjy]

How did you draw the trajectory?Thanks

[xincoder]

@lylyjy Thank you for asking. The visualized results reported in our paper were generated using Plotly. It is very easy to draw it using any library that you are familiar with, e.g., Matplotlib, etc.