Published February 6, 2023
| Version v1
Preprint
Open
Transparent Object Detection and Segmentation in VO Pipeline
Authors/Creators
- 1. Skolkovo Institute of Science and Technology
Description
This paper tests pre-processed data with respect to
deblurring and transparent object segmentation on
traditional pipeline for Visual Odometry i.e. ORB
SLAM 2 and compares with the unprocessed data
on KITTI and TUM sample sequences. We also
tested a well known Deep Neural network by in-
troducing a new coordconv layer to preserve the
spacial information in the pipeline with introduc-
tion of a variable learning rate.