Separating Reflection and Transmission Images in the Wild

12/06/2017
by   Patrick Wieschollek, et al.
0

The reflections caused by common semi-reflectors, such as glass windows, can severely impact the performance of computer vision algorithms. State-of-the-art works can successfully remove reflections on synthetic data and in controlled scenarios. However, they are based on strong assumptions and fail to generalize to real-world images---even when they leverage polarization. We present a deep learning approach to separate the reflected and the transmitted components of the recorded irradiance. Key to our approach is our synthetic data generation, which accurately simulates reflections, including those generated by curved and non-ideal surfaces, and non-static scenes. We extensively validate our method against a number of related works on a new dataset of images captured in the wild.

READ FULL TEXT

page 1

page 2

page 5

page 6

page 8

research
09/30/2021

Fake It Till You Make It: Face analysis in the wild using synthetic data alone

We demonstrate that it is possible to perform face-related computer visi...
research
04/30/2023

Synthetic Data-based Detection of Zebras in Drone Imagery

Datasets that allow the training of common objects or human detectors ar...
research
08/08/2022

Dataset of Industrial Metal Objects

We present a diverse dataset of industrial metal objects. These objects ...
research
02/08/2023

Machine Learning for Synthetic Data Generation: a Review

Data plays a crucial role in machine learning. However, in real-world ap...
research
05/30/2018

CRRN: Multi-Scale Guided Concurrent Reflection Removal Network

Removing the undesired reflections from images taken through the glass i...
research
08/01/2017

PROBE-GK: Predictive Robust Estimation using Generalized Kernels

Many algorithms in computer vision and robotics make strong assumptions ...
research
04/16/2020

In Search of Life: Learning from Synthetic Data to Detect Vital Signs in Videos

Automatically detecting vital signs in videos, such as the estimation of...

Please sign up or login with your details

Forgot password? Click here to reset