carrotsniper/graphtern

Official Code for "A Set of Control Points Conditioned Pedestrian Trajectory Prediction (AAAI 2023)"

A Set of Control Points Conditioned Pedestrian Trajectory Prediction

This repository contains the code for control point conditioned prediction and the initial trajectory refinement network for trajectory prediction.

A Set of Control Points Conditioned Pedestrian Trajectory Prediction
Inhwan Bae and Hae-Gon Jeon
Accepted to AAAI 2023

An illustration of our Graph-TERN model

Graph-TERN Model

• Control points divide each pedestrian's future path into sections and are inferred by each stochastic goal.
• Gaussian mixture model pruning scheme to effectively cutting-off abnormal behaviors of pedestrians.
• Initial trajectory refinement by adding the correction vector field to the linearly interpolated path.

Model Training

Setup

Environment
All models were trained and tested on Ubuntu 18.04 with Python 3.7 and PyTorch 1.6.0 with CUDA 10.1.

Dataset
Preprocessed ETH and UCY datasets are included in this repository, under ./dataset/. The train/validation/test splits are the same as those fond in Social-GAN.

Train Graph-TERN

To train our Graph-TERN on the ETH and UCY datasets at once, we provide a bash script train.sh for a simplified execution.

./scripts/train.sh

We provide additional arguments for experiments:

./scripts/train.sh -p <experiment_tag_prefix> -s <experiment_tag_suffix> -d <space_seperated_dataset_string> -i <space_seperated_gpu_id_string>

# Examples
./scripts/train.sh -d "hotel" -i "1"
./scripts/train.sh -p graph-tern_ -s _experiment -d "zara2" -i "2"
./scripts/train.sh -d "eth hotel univ zara1 zara2" -i "0 0 0 0 0"

If you want to train the model with custom hyper-parameters, use train.py instead of the script file.

python train.py --input_size <input_coordinate_dimension> --output_size <output_gaussian_dimension> \
--n_epgcn <number_of_control_point_gcn_layers> --n_epcnn <number_of_control_point_cnn_layers> \
--n_trgcn <number_of_refinement_gcn_layers> --n_trcnn <number_of_refinement_cnn_layers> \
--n_ways <number_of_control_points>  --n_smpl <number_of_samples_for_refine> --kernel_size <kernel_size>\
--obs_seq_len <observation_length> --pred_seq_len  <prediction_length> --dataset <dataset_name> \
--batch_size <minibatch_size> --num_epochs <number_of_epochs> --clip_grad <gradient_clipping> \
--lr <learning_rate> --lr_sh_rate <number_of_steps_to_drop_lr> --use_lrschd <use_lr_scheduler> \
--tag <experiment_tag>

Model Evaluation

Pretrained Models

We have included pretrained models in the ./checkpoint/ folder.

Evaluate Graph-TERN

To evaluate our Graph-TERN at once, we provide a bash script test.sh for a simplified execution.

./scripts/test.sh -p <experiment_tag_prefix> -s <experiment_tag_suffix> -d <space_seperated_dataset_string> -i <space_seperated_gpu_id_string>

# Examples
./scripts/test.sh
./scripts/test.sh -d "hotel" -i "1"
./scripts/test.sh -p graph-tern_ -s _experiment -d "zara2" -i "2"
./scripts/test.sh -d "eth hotel univ zara1 zara2" -i "0 0 0 0 0"

If you want to evaluate the model individually, you can use test.py with custom hyper-parameters.

python test.py --tag <experiment_tag> --n_samples <number_of_multimodal_samples>

# Examples
python test.py --tag graph-tern_eth_experiment
python test.py --tag graph-tern_hotel_experiment
python test.py --tag graph-tern_univ_experiment
python test.py --tag graph-tern_zara1_experiment
python test.py --tag graph-tern_zara2_experiment

Citation

If you find this code useful for your research, please cite our papers :)

DMRGCN (AAAI'21) | NPSN (CVPR'22) | GP-Graph (ECCV'22) | Graph-TERN (AAAI'23)

@article{bae2023graphtern,
  title={A Set of Control Points Conditioned Pedestrian Trajectory Prediction},
  author={Bae, Inhwan and Jeon, Hae-Gon},
  journal={Proceedings of the AAAI Conference on Artificial Intelligence},
  year={2023}
}

More Information (Click to expand)

@article{bae2021dmrgcn,
  title={Disentangled Multi-Relational Graph Convolutional Network for Pedestrian Trajectory Prediction},
  author={Bae, Inhwan and Jeon, Hae-Gon},
  journal={Proceedings of the AAAI Conference on Artificial Intelligence},
  year={2021}
}

@inproceedings{bae2022npsn,
  title={Non-Probability Sampling Network for Stochastic Human Trajectory Prediction},
  author={Bae, Inhwan and Park, Jin-Hwi and Jeon, Hae-Gon},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year={2022}
}

@inproceedings{bae2022gpgraph,
  title={Learning Pedestrian Group Representations for Multi-modal Trajectory Prediction},
  author={Bae, Inhwan and Park, Jin-Hwi and Jeon, Hae-Gon},
  booktitle={Proceedings of the European Conference on Computer Vision},
  year={2022}
}