Skip to content

Latest commit

 

History

History
29 lines (20 loc) · 999 Bytes

README.md

File metadata and controls

29 lines (20 loc) · 999 Bytes

Automated Generation of Transformations to Mitigate Sensor Hardware Migration in ADS

Learning transformations between vision datasets to overcome sensor hardware versioning

paper: IEEE Robotics and Automation Letters (RA-L) video: YouTube

Overview

The main idea of this project is to come up with a universal transformation between dataset distributions. The difference between these dataset distributions are a result of hardware versioning between sensors. This transformation should preserve several properties of the original dataset:

  1. The prediction of the base model.
  2. The structure and features present in the image.

camera versioning

Problem setup

PreFixer input output

Input and output of trained PreFixer

Training

pip install -r requirements.txt
cd training
python train_lenscoder.py <path-to-dataset>