search

m-hasan-n / pooling

"Pooling Toolbox" is the code of our work "Maneuver-Aware Pooling for Vehicle Trajectory Prediction".
MIT License
50 stars 8 forks source link

Poor performance on longitudinal maneuver inference #6

Open harryguantj opened 7 months ago

harryguantj commented 7 months ago

As follows, the accuracy of the longitudinal maneuver is poor, and after trying to change how we divide and tag the longitudinal maneuvers, the accuracy is volatile when the threshold changes. I'm crious about this phenomenon and hope for more discuss.

harryguantj commented 7 months ago

This could be confirmed in the original CSP or so-called Convolutional Social Pooling LSTM.

harryguantj commented 7 months ago

Change the parameter 'acc_threshold' to another one instead of 0.7, the accuracy of the longitudinal maneuver will perform better. As shown in the picture below, the accuracy performs better and the RMSE of all types decreases. 微信图片_20240204213448 微信图片_20240204213500

CrazyWangBa commented 6 months ago

Change the parameter 'acc_threshold' to another one instead of 0.7, the accuracy of the longitudinal maneuver will perform better. As shown in the picture below, the accuracy performs better and the RMSE of all types decreases. 微信图片_20240204213448 微信图片_20240204213500

I'm also experiencing this problem when trying the code for this repository, may I ask your setting of acc_threshold?

guoyage commented 5 months ago

Change the parameter 'acc_threshold' to another one instead of 0.7, the accuracy of the longitudinal maneuver will perform better. As shown in the picture below, the accuracy performs better and the RMSE of all types decreases. 微信图片_20240204213448 微信图片_20240204213500

I'm also experiencing this problem when trying the code for this repository, may I ask your setting of acc_threshold?

I found that the acceleration threshold in the original paper was 0.2, but I tried changing this value and the RMSE did not improve much.