VectorNet: Encoding HD Maps and Agent Dynamics from Vectorized Representation, partial function re-implement

Main model is saved in 'GlobalGraph.py', 'SubGraph.py' and 'MLP.py'. 'generateH5.py' and 'run_All_Test.py' are used for generating model's answer on Argoverse test dataset, then submit to Argoverse Motion Forecasting Competition link to https://evalai.cloudcv.org/web/challenges/challenge-page/454/overview.

Argoverse dataset works on Ubuntu or MacOS, however this project implemented on Windows, so you should do a simple work to run, which means you can transform this project to other platforms or save data to root dir then run 'Main.py'.

Lots of '.model' files are some results generated by different dataset or few changes on model, you can directly use them. 'visualization.py' can give a naive visualization result, you also can use Argoverse's visualization.

If you can read Chinese, welcome to this link for more details, you will see some valuable discussion and some visualization pictures!