LaneMarkingDetectionAssessment - Silent Testing

This repo aims for evaluating a lane marking detection algorithm based on the Silent Testing concept.

Setup:

• Creating a conda environment, and install the requirements.txt.
• Download the NuScene Dataset and also install the nuscenes-devkit.
• The dataset and the devkit should be put at the same level of directory of the main code.

Run

• using the GetResolution.ipynb to get the pixel per meter in x and y directions.
• Enter the above values in LaneDetection.py, which also need the source points as input.
• You can use the FindScene.py to find the scene and sample id given an image name

Adversarial images The code to generate adversarial images from large-scale open-sourced datasets can be found here.