It is a model that predicts the trajectory of the Vehicle by using deep learning.
We used Attention, LSTM, etc. to solve various problems of existing MTP. The path is generated for all the agents on the image, not for one agent.
If the train or test (visualization) is to be carried out smoothly, the form of the directory should be as shown in the figure above. When the git was first clone, there would not be an experiment and data folder, which was uploaded separately using drive because of its large capacity. It is important to note that the data folder should be unzip so that there is no additional data folder in the data folder.
https://drive.google.com/file/d/1nf8w2YCTlLGRb_HbrB6cW1TXp3rEJoq6/view?usp=sharing
You can download dataset through link. We recommend using the dataset we provide because we used the nuscene dataset by processing. The size of the dataset is so large that it is recommended to run with enough space. Compressed file is 100GB.
https://drive.google.com/file/d/18MgeVviFO9e4zpctcg3T314BQ_6--2t5/view?usp=sharing
Because the trained model also has a large capacity, you need to download it through drive link. Directory is created by unziping a file posted on the drive by a folder called 'experiment', as shown in the picture above. If you unzip just like the directory picture, it will work without any problems.
The code was written using python 3.6, with Linux environment. For the convenience of users, the conda environment has been copied. If you use this environment as it is, you will be able to use it without having to download additional libraries. Since we have done so in the U environment, we recommend that you do so in the same environment.
conda env create -f CS470_environment.yml
conda activate new_py
Or you can set up with pip by running:
pip install -r requirements.txt
You can train with simple commands like the following by setting the required argumentss on 'main.py' in advance.
python3 main.py
We stored about 80 printed images in '/test/results' in advance. However, if you want to print a new images and see the ADE, FDE results, you can see the new 80 imagess and the following results with the command below.
python3 test_visualize.py
