RCAR2021_MultiAgentTrajPrediction

code and materials for Multi-Agent Trajectory Predition Based on Graph Neural Network

Note:This repo stops to be updated, and for further work of us, you can switch to this repo which keeps on updating.

Heterogeneous GNN

• main.py: main function of LOG Analysis model training process

• parameters.py: parameters management of the training process

• dataFormat.py: basic data format of player, ball and game data

• referee.py: referee command defined in proto file

• dataPreprocess.py: preprocess the text file we get from our vision module, converting into formatted data we can use and doing Min-Max normalization

• SSLDataset.py: construct the graph structure for future gnn training

• mys2v.py: basic graph neural network we use

• pna/mypna.py: Heterogeneous PNAConv

• heterogeneous/myheter.py: Heterogeneous graph neural network frame supporting different GNNs

• Net.py: neural network we construct

• debug/debug.py: draw gradient of graph neural network

• visualize.py: draw pictures of our training result

• testOurModel.py: load torch model and see results and also visualiztion

Homogeneous GNN