/Multi-Class-Social-STGCNN

Building on the work of Social STGCNN we introduce Multi-Class Social STGCNN to include class information into trajectory prediction to improve the accuracy.

Primary LanguagePython

Video_Trajectory_Predictor

Installation

install requirements $ pip install -r requirements.txt

Train

run train.py

Test

run test.py

Visualise

run videoStreaming.py and go to localhost:5000

Config

path - location of video or training data

class_enc - True or False whether to include class encoding module in model

class_weighting - when training weight loss based on class weight

frameSkip - number of frames to skip between coordinate samples

labels - class types to train/test on

checkpoint - location of pretrained model

percentageToRemove - remove edge data from training data

showGroundTruth - show ground truth vs predictions on visualisations

saveImages - save ground truth vs predictions frames when a new prediction is made