This repository is the re-implementation of GRIP++: Enhanced Graph-based Interaction-aware Trajectory Prediction for Autonomous Driving on the Baidu Apollo Trajectory dataset. GRIP++ is an enhanced version of GRIP (GRIP: Graph-based Interaction-aware Trajectory Prediction).
The official codes can be found here GRIP++.
Our implementation has several advantages over official codes:
-
User friendly graph data preprocessing.
-
Using graph convolutional network rather than Conv2d: The official codes implemented GCN using Conv2d.
-
Better performance: Our model has better performance than the official implementation under same hyperparameters.
| WSADE | ADEv | ADEp | ADEb | |
|---|---|---|---|---|
| Our implementation | 1.133 | 1.705 | 0.696 | 1.762 |
| official implementation | 1.176 | 1.652 | 0.764 | 1.829 |
- data generation.
python data_process.py- training model.
python main.py