search

agrimgupta92 / sgan

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

Evaluation error for model without pooling module (KeyError : 'g_state') #55

Open iamrjkunal opened 5 years ago

iamrjkunal commented 5 years ago

Hi I have trained the model for zara1 (8 timestamps) by using the pretrained model parameters (which I got by using print arguments of your pretrained model). Evaluation script is not working in case of no pooling module whereas it works perfectly for pooling part. Screenshot from 2019-06-23 12-12-56 Can you help me out? @agrimgupta92 Thank you in advance P.S: Anyone who resolved the same error, please help me :)

gwiplot commented 5 years ago

I had the same problem. Did you solve it?

ChaofanTao commented 5 years ago

I had the same problem. Did you solve it?

From the train.py line 344, we can observe that some of the keys like "g_state" is blocked in the "checkpoint_no_model.pt". I solve the problem by just running the "checkpoint_with_model.pt". You can operate it by simply deleting the "checkpoint_no_model.pt"

xjsxjs commented 2 years ago

Hi, I have a question, why use "g_state" instead of "g_best_state" in evaluate, hope to get your help. Thanks.