search

nachiket92 / conv-social-pooling

Code for model proposed in: Nachiket Deo and Mohan M. Trivedi,"Convolutional Social Pooling for Vehicle Trajectory Prediction." CVPRW, 2018
MIT License
336 stars 102 forks source link

evaluate.py #21

Closed beyond96 closed 3 years ago

beyond96 commented 4 years ago

Hi, In the 'evaluate' program, what are the meanings of the two parameters, 'in_length = 16' and 'out_length = 25'? Don't we want to predict the vehicle trajectory in the next 5s? Why 'out'_length = 25?

kshitij002 commented 4 years ago

hi @beyond96 were you able to crack why the out_length = 25, even though we have to predict for the next 5 seconds?? Also the evaluate.py outputs the tensor of size 25, could you help us @nachiket92 ??

nachiket92 commented 4 years ago

The sampling rate used is 5 Hz. Thus the input is the past 15 locations + the current location (3 sec history), and the prediction has 25 locations corresponding to 5 sec.

kshitij002 commented 4 years ago

how did you deduced the result table reported in publication, from the tensor with prediction 25 location for corresponding to 5 sec? Looking forward to hear from you soon. thanks for your valuable time :)

nachiket92 commented 4 years ago

That would be the last (25th) value of the tensor

kshitij002 commented 4 years ago

I am so sorry I guess I couldn't express my doubts to you in a clear manner. I am concerned about the results reported in the table, how did you report the values of NLL/RMSE for all the (t+5) seconds from the tensor of 25 size. Looking to hear from you soon, thanks for your valuable time!! oie_y21TSecIeemb

DataDemystifier commented 4 years ago

It is because of the sampling rate of 5 Hz like was already mentioned by the author of the paper and repository. The tensor has length 25 and therefore the values correspond to the following sequence of times: [0.2s, 0.4s, 0.6s, 0.8s, 1.0s, ... , 4.8s, 5.0s]

kshitij002 commented 4 years ago

@DunkDream Thanks a lot

rrryan2016 commented 2 years ago

For data in NGSIM, the shape of both fut and fut_pred is shape of fut: torch.Size([25, 32, 2]). Regarding to the conversation above, 25 means the following 5 second, but what does 32 mean? Thanks in advance