Subsea Inpainting Dataset

The Subsea Dataset comprises small clips of subsea inspection videos with masks covering overlays. This dataset is designed to support the Oil and Gas industry, which generates thousands of hours of subsea inspection videos daily and holds millions of hours of archived footage.

Context

In the Oil and Gas industry, underwater videos are typically captured by Remote Operated Vehicles (ROVs). These videos are integral to the integrity management of subsea assets, such as pipelines and Wet Christmas Trees (WCTs). They enable Oil and Gas companies to maintain the safety and efficiency of their offshore operations.

A common feature in these videos is a “burned overlay” that displays real-time metadata, including ROV coordinates, water depth, and heading. While these overlays ensure that metadata remains attached to the visual data even after manipulations like clipping videos or extracting frames, they can also obstruct critical image information and confuse image processing algorithms.

This dataset aims to test the capabilities of current state-of-the-art inpainting techniques in removing overlays from subsea inspection videos.