search

agrimgupta92 / sgan

Code for "Social GAN: Socially Acceptable Trajectories with Generative Adversarial Networks", Gupta et al, CVPR 2018
MIT License
819 stars 260 forks source link

Unable to get decent performances upon training the model #6

Closed AyeshaNirma closed 6 years ago

AyeshaNirma commented 6 years ago

Hi Agrim,

I have used default parameters to train on zara01 as described in read me. The errors are horrendous so for pred len : 8 and 12 the ADE is 2.27 and 9.02 respectively. Its a but puzzling as I have done absolutely no changes to your code or model whatsoever and I have used all default params. Could you perhaps point out what potentially could be the issue.

agrimgupta92 commented 6 years ago

When you say default parameters there are 3 possibilities:

  1. Default params in the train script
  2. Params in run_traj.sh script
  3. Params from running print args on one of the pre-trained models. I would trust 2,3 more. Training GANs in my experience was really tough and the hyper-parameters in 2,3 worked well for me. I would suggest looking at weight init.