Semester project of Master of Computer Science in EPFL
Student name : Baran Nama
Advisor: Alexandre Alahi
Presentation : https://drive.google.com/file/d/1biC23s1tbsyDETKKBW8PFXWYyyhNEAuI/view?usp=sharing
Baseline implementation: https://github.com/vvanirudh/social-lstm-pytorch
Paper: http://cvgl.stanford.edu/papers/CVPR16_Social_LSTM.pdf
Detailed info about challange and datasets: https://www.aicrowd.com/challenges/trajnet-a-trajectory-forecasting-challenge
Made improvements: Please see attached presentation
- generator.py : Python script for generating artifical datasets
- helper.py: Python script includes various helper methods
- hyperparameter.py: Pyton script for random best parameter selection for a model
- make_directories.sh: Bash script for creation of file structure
- model.py: Python file includes Social LSTM model definition
- olstm_model.py: Python file includes Occupancy LSTM model definition
- olstm_train.py: Python script for training Occupancy LSTM model
- test.py: Python script for model testing and getting output txt file for submission
- train.py: Python script for training Social LSTM model
- utils.py: Python script for handling input train/test/validation data and batching it
- validation.py: Python script for externally evaluate a trained model by getting validation error
- visualize.py: Python script for visualizing predicted trajectories during train/test/validation sessions
- vlstm_model.py: Python file includes Vanilla LSTM model definition
- vlstm_train: Python script for training Vanilla LSTM model
- Fork the repository
- Start train a model >>> python train/olstm_train/vlstm.train.py - -[Parameter set]
- If necesarry file structure is not exist (which is the initial situation), train script will run make_directories.sh and this command will automatically create file structure
- Enjoy!
| Model name | Avarage error | Final error | Mean error |
|---|---|---|---|
| Social LSTM | 1.3865 | 2.098 | 0.675 |
| Occupancy LSTM | 2.1105 | 3.12 | 1.101 |
| Vanilla LSTM | 2.107 | 3.114 | 1.1 |
Reference: http://trajnet.stanford.edu/result.php?cid=1