Get our free extension to see links to code for papers anywhere online!

Chrome logo Add to Chrome

Firefox logo Add to Firefox

A Generative Approach for Detection-driven Underwater Image Enhancement

Dec 10, 2020
Chelsey Edge, Md Jahidul Islam, Christopher Morse, Junaed Sattar

In this paper, we introduce a generative model for image enhancement specifically for improving diver detection in the underwater domain. In particular, we present a model that integrates generative adversarial network (GAN)-based image enhancement with the diver detection task. Our proposed approach restructures the GAN objective function to include information from a pre-trained diver detector with the goal to generate images which would enhance the accuracy of the detector in adverse visual conditions. By incorporating the detector output into both the generator and discriminator networks, our model is able to focus on enhancing images beyond aesthetic qualities and specifically to improve robotic detection of scuba divers. We train our network on a large dataset of scuba divers, using a state-of-the-art diver detector, and demonstrate its utility on images collected from oceanic explorations of human-robot teams. Experimental evaluations demonstrate that our approach significantly improves diver detection performance over raw, unenhanced images, and even outperforms detection performance on the output of state-of-the-art underwater image enhancement algorithms. Finally, we demonstrate the inference performance of our network on embedded devices to highlight the feasibility of operating on board mobile robotic platforms.

* Under review for ICRA 2021 

Share this with someone who'll enjoy it:

   Access Paper Source

Share this with someone who'll enjoy it: