Human Trajectory Prediction via Counterfactual Analysis (CausalHTP)
The official PyTorch code implementation of "Human Trajectory Prediction via Counterfactual Analysis" in ICCV 2021.arxiv
News
We add the implementation of our project Causal-STGAT, where we apply our CausalHTP method to the baseline backbone network STGAT. The code of Causal-STGCNN is coming soon.
Introduction
Most trajectory prediction methods concentrate on modeling the environment interactions and aggregate these interaction clues with history behavior clues for trajectory prediction. However, there are heavy biases in the between training and deployment environment interactions. The motivation of this project is to mitigate the negative effects of the inherent biases. We propose a counterfactual analysis method to alleviate the overdependence of environment bias and highlight the trajectory clues itself. This counterfactual analysis method is a plug-and-play module which can be easily applied to any baseline predictor, and consistently improves the performance on many human trajectory prediction benchmarks.
Figure 1. Training process of our counterfactual analysis method. We apply the counterfactual intervention by replacing the features of past trajectory with the counterfactual features such as uniform rectilinear motion, mean trajectory, or random trajectory. The counterfactual prediction denotes the biased affect from environment confounder. To alleviate the negative effect of environment bias, we subtract the counterfactual prediction from original prediction as the final causal prediction.
Requirements
- Python 3.6+
- PyTorch 1.3
To build all the dependency, you can follow the instruction below.
pip install -r requirements.txt
Dataset
The datasets can be found in datasets/, we provide 5 scenes including eth, hotel, univ, zara1, and zara2.
Training and Evaluation
You can train the model for eth dataset as
python train.py --dataset_name eth
To evaluate the trained model, you can use
python evaluate_model.py --dataset_name eth --resume your_checkpoint.pth.tar
The pre-trained models can be found in pretrain/
Result
| Results (ADE/FDE) | ETH | HOTEL | ZARA1 | ZARA2 | UNIV | AVG |
|---|---|---|---|---|---|---|
| STGAT | 0.73/1.39 | 0.38/0.72 | 0.35/0.69 | 0.32/0.64 | 0.57/1.22 | 0.47/0.93 |
| Causal-STGAT | 0.60/0.98 | 0.30/0.54 | 0.32/0.64 | 0.28/0.58 | 0.52/1.10 | 0.40/0.77 |
Figure 2. Visualization examples of our Causal-STGAT method and baseline Social-STGAT method in the different scenes in the both ETH and UCY datasets. The comparisons quantitatively demonstrate the effectiveness of our counterfactual analysis on the RNN-based baselines.
Citation
Part of the code comes from STGAT. If you find this code useful then please also cite their paper.
Please use the citation provided below if this repo is useful to your research:
@inproceedings{CausalHTP,
title={Human Trajectory Prediction via Counterfactual Analysis},
author={Chen, Guangyi and Li, Junlong and Lu, Jiwen and Zhou, Jie},
booktitle={ICCV},
year={2021}
}