search

abduallahmohamed / Social-STGCNN

Code for "Social-STGCNN: A Social Spatio-Temporal Graph Convolutional Neural Network for Human Trajectory Prediction" CVPR 2020
MIT License
480 stars 140 forks source link

Details about data processing #64

Open guilunmeibeautiful opened 1 year ago

guilunmeibeautiful commented 1 year ago

Great work, thanks for sharing the code. I am a novice in the field of trajectory prediction. I read the code and encountered some problems that I don't understand. Please help me answer them.

  1. What is "Make coordinates relative"?
  2. Why is there a distinction between "Linear vs Non-Linear Trajectory"?
  3. What is the function of the poly_fit method in utils.py? Thank you very much!
guilunmeibeautiful commented 1 year ago

Also, how to train and test the code under windows?

abduallahmohamed commented 1 year ago

Thanks for asking. This data loader is taken from Social-GAN. 1- Relative to the first position, they are also delta between position aka speed 2- This is a very old topic, but lines trajectories are easier to predict using classical methods and models like constant velocity models. On the other hand, non-linear ones have patterns that are hard to model like sudden change in trajectory or sudden stop. 3- It just fits a polynomial to the trajectory, very old approach to benchmark against showing the benefit of data driven models. I encourage to take a look into our latest work "Social-Implicit".

abduallahmohamed commented 1 year ago

I'm not sure about windows as I never used it before for development.