trungmanhhuynh / 3d_human_trajectory_prediction

Human-Trajectory-Prediction

Using 2d-grid scene lstm with hard and soft filter to improve individual human trajectory prediction.

Prerequisites

• Pytorch
• Python 3.4

Before Training/Testing Your Models

Run following command to create directories for each model

>>  sh make_dir.sh               

Because we train each model in 2 stages, so we also need to create directories for stage 2 of each model. In make_dir.sh, change {model_name}_stage_1 to {model_name}_stage_2 and run above command again. Applying the same concept, you can also create directories for other models/configurations.

Test Using Pre-trained Model

Train Your Own Models:

 python train.py --model_dataset 4 --train_dataset 0 1 2 3 --use_scene --use_distance --nonlinear_grids

Please read the config.py for further details of each flag.