search

realcrane / Human-Trajectory-Prediction-via-Neural-Social-Physics

Our ECCV 2022 paper Human Trajectory Prediction via Neural Social Physics
127 stars 19 forks source link

How to train the 2nd model? #9

Open Sigrid-Z opened 1 year ago

Sigrid-Z commented 1 year ago

Thank you for your work~

I have trouble training the 2nd model, since the semantic masks are not compatible with the train pickle data? Could you explain how to get the semantic masks data? It seems the preprocess code is incomplete.

Waiting for your answer, thx. image

JiangbeiYue commented 1 year ago

Hi,

For this problem, you can change the 274 line and 276 line in train_nsp_wo.py into: semantic_maps_name_train = sorted(os.listdir(semantic_path_train)) semantic_maps_name_test = sorted(os.listdir(semantic_path_test)). This might be because It is different to read files for different computers or IDEs.

The semantic masks are generated by the goal sampling network detailed in our supplementary material which is similar to the goal prediction network in Y-net.