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
326 stars 97 forks source link

preprocess_data.m #20

Closed beyond96 closed 4 years ago

beyond96 commented 4 years ago

The program with 6 workers has been running for 40 hours and still hasn't finished. What's wrong?

Xiaoyu006 commented 4 years ago

Hi, Just finished data preprocessing using provided .m file.

It took me about 36 hours to finish the preprocessing with 6 best i9 workers.

And finally I got three .mat files as expected. the size of these .mat files are: TestSet.mat - 27.3MB TrainSet.mat - 107.6MB ValSet.mat - 15.6MB

beyond96 commented 4 years ago

Thanks for your answer！ I have another question, how to get the predicted trajectory distribution of vehicles from your program？In other words, which variable in the program records the predicted trajectory distribution？fut_pred？ Many thanks!

kshitij002 commented 4 years ago

@beyond96 how much time did it took for complete preprocessing of the data?? and did you implemented the paper using authors code?? were the results obatained same as they were reported in the paper??

Regards kshtiij

thelastpolaris commented 4 years ago

Can somebody share the resulting preprocessed data? Many thanks!

@Xiaoyu006 could you kindly share the resulting files?

thelastpolaris commented 4 years ago

The preprocessing is done in an extremely inefficient way. I made a pull request and in case it is not accepted you can grab the improved code here https://github.com/thelastpolaris/conv-social-pooling/blob/fix_preprocessing/preprocess_data.m

With this script preprocessing can be done in 20 minutes

Xiaoyu006 commented 4 years ago

@thelastpolaris Thanks for your work! Could you please add more comments to you matlab code? I am new to matlab.

Regards.

thelastpolaris commented 4 years ago

@Xiaoyu006 the only difference is that on each iteration instead of going through the whole dataset looking for vehtraj that corresponds to vehicle id of a current frame and doing the same thing for trajectories that correspond to the current timeframe I precompute these trajectories into two dictionaries (Maps). Then you don't need to go through the whole dataset but just through vehTrajs and vehTimes

Xiaoyu006 commented 4 years ago

@thelastpolaris Great idea! And thanks for your timely reply.

nachiket92 commented 4 years ago

Thanks @thelastpolaris! Running your script right now, will merge.

CuttlefishXuan commented 3 years ago

Can somebody share the raw txt data? Thanks.