search

abduallahmohamed / Social-STGCNN

Code for "Social-STGCNN: A Social Spatio-Temporal Graph Convolutional Neural Network for Human Trajectory Prediction" CVPR 2020
MIT License
483 stars 141 forks source link

Question about evaluation #27

Closed HRHLALALA closed 4 years ago

HRHLALALA commented 4 years ago

Hi, thanks for your good work. I have a question regarding the evaluation. I notice that you follow the steps like:

  1. calculate the FDE and ADE 20 times for each pedestrian
  2. choose the min one
  3. calculate the average value image

However, the evaluation in social gan follows the way like:

  1. sum up the ADE and FDE scores for all pedestrians in a single scene
  2. repeat 20 times
  3. choose the min one.
  4. repeat the steps above and average the scores through all scenes. image

Do you think that these two different processes make the evaluation unfair? Please correct me if I misunderstand these two steps. I hope I can get your answer soon.

abduallahmohamed commented 4 years ago

Hi, Thanks for asking! Please do check this https://github.com/abduallahmohamed/Social-STGCNN/issues/14#issuecomment-605328659 and let me know if your question still unanswered after this.

HRHLALALA commented 4 years ago

Thanks for providing this comment. Maybe it is dumb to ask. Would you please give a deeper explanation of "distribution" and "expectation" you mentioned?

abduallahmohamed commented 4 years ago

Thanks for asking again. Our model predicts a bi-variate Gaussian distribution per each time step per pedestrian. Then later on we sample these distributions to generate the trajectories. these distributions become uncorrelated in the sampling unlike Social-GAN where it generates a sample of trajectories( expectation) and do this 20 times. Let me know if this still unclear On Sun, Jun 28, 2020 at 3:26 AM HRHLALALA notifications@github.com wrote:

Thanks for providing this comment. Maybe it is dumb to ask. Would you please give a deeper explanation of "distribution" and "expectation" you mentioned?

— You are receiving this because you commented.

Reply to this email directly, view it on GitHub https://github.com/abduallahmohamed/Social-STGCNN/issues/27#issuecomment-650717073, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFIVLG7FMW4EICVHR75BH73RY35B5ANCNFSM4OKLW27A .

--

Abduallah Mohamed abduallahmohamed.com

HRHLALALA commented 4 years ago

Thanks. It is clear now