Constraints do influence the trained image without using CLIP. E.g., the image mean can be trained to be above or below a specific threshold.
Constraints only affect the pixels that are processed by the transforms of the target.
Here's a list of all available constraints:
- blur: Adds the difference between the image and a blurred version to the training loss.
- border: Adds a border with a specific size and color to the training loss.
- contrast: Pushes the contrast above or below a threshold value.
- edge_mean: Adds the difference between the current image and and an edge-detected version to the training loss.
- mean: Pushes the image color mean above or below a threshold value.
- noise: Adds the difference between the current image and a noisy image to the training loss.
- normalize: Adds image normalization to the training loss.
- saturation: Pushes the saturation above or below a threshold value.
- std: Pushes the standard deviation above or below a threshold value.