We present a fully-convolutional conditional GAN-based model for fast underwater image enhancement, which we refer to as FUnIE-GAN. It is designed for real-time use by visually-guided underwater robots operative in noisy visual conditions. We also present a large-scale dataset named EUVP to facilitate paired and unpaired learning of underwater image enhancement.

 funie_fig2 funie_fig1

FUnIE-GAN can learn to enhance perceptual image quality from both paired and unpaired training. More importantly, the enhanced images significantly boost the performance of several underwater visual perception tasks such as object detection, human pose estimation, and saliency prediction. Detailed results and demonstrations can be found in the paper


In addition to providing state-of-the-art enhancement performance, FUnIE-GAN offers fast inference on single-board computers (e.g., 48+ FPS on Jetson AGX Xavier, 25+ FPS on Jetson TX2), which makes it feasible for real-time robotic applications. Check out this GitHub repository for FUnIE-GAN models and associated training pipelines.  

 Important pointers: