Low Light Video Enhancement using Synthetic Data Produced with an\n Intermediate Domain Mapping

Advances in low-light video RAW-to-RGB translation are opening up the\npossibility of fast low-light imaging on commodity devices (e.g. smartphone\ncameras) without the need for a tripod. However, it is challenging to collect\nthe required paired short-long exposure frames to learn a supervised mapping.\nCurrent approaches require a specialised rig or the use of static videos with\nno subject or object motion, resulting in datasets that are limited in size,\ndiversity, and motion. We address the data collection bottleneck for low-light\nvideo RAW-to-RGB by proposing a data synthesis mechanism, dubbed SIDGAN, that\ncan generate abundant dynamic video training pairs. SIDGAN maps videos found\n'in the wild' (e.g. internet videos) into a low-light (short, long exposure)\ndomain. By generating dynamic video data synthetically, we enable a recently\nproposed state-of-the-art RAW-to-RGB model to attain higher image quality\n(improved colour, reduced artifacts) and improved temporal consistency,\ncompared to the same model trained with only static real video data.\n