These docs updates are addressing #122 . The only point not addressed is a channel attack generation script. It was not included in the repo and renditions generated with it weren't added to the dataset, because such an attack is very trivial, if there's any metric which uses all image channels. Instead of that, a specific unit test and a channel-manipulated rendition video were included in the repo, to make sure all all-channel metrics are not accidentally excluded from models.
These docs updates are addressing #122 . The only point not addressed is a channel attack generation script. It was not included in the repo and renditions generated with it weren't added to the dataset, because such an attack is very trivial, if there's any metric which uses all image channels. Instead of that, a specific unit test and a channel-manipulated rendition video were included in the repo, to make sure all all-channel metrics are not accidentally excluded from models.