课程: Self-Supervised Machine Learning

免费学习该课程！

今天就开通帐号，24,100 门业界名师课程任您挑！

Pretext task: Image colorization

Pretext task: Image colorization

课程: Self-Supervised Machine Learning

开始一个月试用 购买团队方案

Pretext task: Image colorization

“

- [Instructor] Another self-supervised pretext task that can be used to learn image representations is image colorization. Here the task is actually to predict the color of a gray image. Now, you may start off with colored images. Well, getting them in grayscale is very straightforward. You just remove the color from the input image and feed in the gray image to the model and have it predict its color. The idea behind this task is that your model needs to learn some meaningful features and meaningful representations of the underlying data in order to make its color predictions. For example, if you need to know that the sky is blue, tree leaves are green, the sand is brown, if there is mud, it's brown, if there is water, maybe blue again. So you start off with the raw unlabeled data. These are all multi-channel images, that is color images. You then remove the color from these images, you apply a grayscale…

内容