# VECTORNET_LSTM_EVALUATOR

# Introduction

Vectornet lstm evaluator is is based on a GNN encoding network taking environments and traffic agents as nodes. Comparing with semantic map CNN model, it has less parameters to learn and better performance.

![Diagram](images/vectornet.svg)

# Where is the code

Please refer the online inference [vectornet lstm evaluator](https://github.com/ApolloAuto/apollo/modules/prediction/evaluator/vehicle/vectornet_evaluator.h).

## References

[1]: Gao, Jiyang, et al. "Vectornet: Encoding hd maps and agent dynamics from vectorized representation." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2020.