Deep Learning Approach to Global Bird's Eye View Semantic Mapping

Setup

1. Clone the repo
2. Download the AI-IMU dataset from ai-imu's dropbox and store in root directory.
3. Download the local BEV data from Dropbox
4. Store MotionNet_Prediction and MotionNet_Prediction_Test folders from the data in root directory.
5. Create temp and results folder under root directory.
6. Set the working directory to root.
7. The root directory should look like this:
   
8. Run python3 kitti_mapping_opt.py
9. If you want to run main_KITTI.py, please call it under root directory, and run python3 ai_imu_dr/src/main_kitti.py

Motion Net

1. Create a conda environment with all the dependicies to run MotionNet by running the following line in 'motionnet' dir:

conda env create -f environment.yml
conda activate motionnet

2. TODO

Evaluation

We use evaluation matrics including IoU, precision, recall to evaluate the proformance.

TODO:

1. Run python3 kitti_mapping_opt.py to get the .npy files (in root directory).
2. Move the generated .npy files to evaluation/ folder.
3. Run 'python3 evaluate.py', change the path in evaluate.py if necessary.