Open arstropica opened 1 year ago
For anyone else that is interested in how to transform 3-channel videos into a suitable data format for npy, check out the first referenced article: Learning to See in the Dark. P. 4 describes the general method using bayer images and black levels.
I came up with this script to convert videos to the npy format, but it appears as though the input data needs to be raw sensor data and not RBG video, so the results are disappointing.
If there is a pre-processing pipeline for preparing a for video for testing, it could be feasible to use this script to improve low exposure videos. As-is, without raw input, the end product is unusable.
What do the values in the 4th dimension of the npy array represent? Is there a way to generate a base npy from custom videos for training?