data process

[zhaolongtong]

Why are the candidate trajectories processed like this?

The reason is that the recent blocks are filtered incorrectly, as shown in the figure below. The upper left image is the nearest block to the vehicle, the upper right image is the Bezier curve generated from the vehicle position to the nearest block, the lower left image is the filtered paths , and the lower right image is the final Scene visualization.

[Hailan-9]

Hello brother, may I ask if you can successfully train this DTPP code? I couldn't train successfully on my end, and the error I reported was a problem with reverse differentiation, loss. backward(). I debugged it and found that it was in the scenario_tree_prediction. py file with enw_decode=self. microenvironment_decoder (query, encoding, encoding, env-mask)
And ego_cndition_decoding=self. ego_dedition_decoder (query, At the locations of ego_traj_decoding, ego_traj_decoding, and ego_cndition_mask, the solved decoding has a value of 'nan'. May I ask if yours is also like this. My environment configuration should be consistent with what is written in DTPP's readme. If it's convenient to reply, could you please reply to my email [email protected] .
Thank you, brother. Looking forward to hearing from you!

[Hailan-9]

Hello brother, may I ask if you can successfully train this DTPP code? I couldn't train successfully on my end, and the error I reported was a problem with reverse differentiation, loss. backward(). I debugged it and found that it was in the scenario_tree_prediction. py file with enw_decode=self. microenvironment_decoder (query, encoding, encoding, env-mask) And ego_cndition_decoding=self. ego_dedition_decoder (query, At the locations of ego_traj_decoding, ego_traj_decoding, and ego_cndition_mask, the solved decoding has a value of 'nan'. May I ask if yours is also like this. My environment configuration should be consistent with what is written in DTPP's readme. If it's convenient to reply, could you please reply to my email [email protected] . Thank you, brother. Looking forward to hearing from you!

The problem has been resolved. I mistakenly converted the passed atte_mask type to boolean type, which should be tensor.float type.