search

agrimgupta92 / sgan

Code for "Social GAN: Socially Acceptable Trajectories with Generative Adversarial Networks", Gupta et al, CVPR 2018
MIT License
820 stars 261 forks source link

Does the parameter "num_epochs" really work? #82

Open a79687417 opened 4 years ago

a79687417 commented 4 years ago

I set "num_epochs" 200 as default, but the training process usually goes up to 350+. It seems like, in the code, this parameter is only for counting total iterations, and does nothing in actual iterating.

Jacobieee commented 3 years ago

I think it depends on args.num_iterations. So, in line 123, 125 and 227 in train.py, it calculated the number of iterations again which is determined by how large is your dataset, batche_size and d_steps.