Discrete3DOD
Official repository for IEEE Transaction on Intelligent Transportation System, entitled "One Stage Monocular 3D Object Detection Utilizing Discrete Depth and Orientation Representation."
Requirements
- Python 3.7
- Pytorch 1.9
- OpenCV > 3.0
- Tqdm
- Cityscapesscripts (https://github.com/mcordts/cityscapesScripts)
- PyQuaternion
Configure the settings
Make changes to the settings here:
- Change the dataset root
- Change weights name/location
- Change the location of the save folder for saving prediction results
- Download our pretrained weights here
Running Evaluation
To run the cityscapes benchmark simply
python evaluate.py
The results will be saved in saves/results/results.json
Visualize Results
python gen_vis.py
The results will be saved in folder saves/vis/