Sprout a Lens module #5
Conversation
See: #4 As we learn how to Swift; we're stepping away from the SwiftUI/etc. system and focusing more on building a library for doing the filtering. We'll start with downloading a file from the Internet, saving it, then running it through a LensStack, and saving a copy at each stage. This way, we can start to diagnose if/how the filters are working without futzing about with SwiftUI; which adds a confounding factor when trying to figure things out.
There was a problem hiding this comment.
Annnnnd it look like the fine folks at Apple already thought of 90% of this little sub-project: Recipes for Custom Effects
You can create custom effects by using the output of one image filter as the input of another, chaining as many filters together as you’d like. When you create an effect this way that you want to use multiple times, consider subclassing CIFilter to encapsulate the effect as a filter.
|
@theInternauts That's cool! I think a Swift Wrapper may still make sense for that functionality. However, it's up to you to decide what path you want to take! |
…ilters and added a test filter to the driver code--currently unable to save filtered image
|
|
||
| func apply(image: CIImage) -> CIImage { | ||
| let localFilter = CIFilter(name: filter.filterName, parameters: filter.filterData) | ||
| localFilter?.setValue(image, forKey: kCIInputImageKey) |
There was a problem hiding this comment.
This looks great! It feels a little bit like feature envy for the Lens to be pulling this data out of the filter. I wonder if we can apply "Tell, Don't Ask" here and refactor in such a way that the Lens method takes responsibility for handling the failure case for Filters (i.e. the filter raises, returns nil, etc.).
guard let output = filter?.apply(image, fingerprint) else {
return image
}
return output
This way each filter can encapsulate how it leverages the fingerprint to generate the filter parameters.
For example, a MotionBlur filter looks like it wants two numbers between ??? and 20?, so the MotionBlur filter class would use the Fingerprint to generate two random values. On the other hand, the SepiFilter class would only need to generate a single float value; so it would use the fingerprint to do that. Her'es some pseudocode for those two filters:
class SepiaFilter {
func apply(image, fingerprint) {
intensity = fingerprint.randBetween(0.8,1.0)
let localFilter = CIFilter(name: "CISepiaTone", parameters: [kCIInputIntensityKey: intensity])
localFilter?.setValue(image, forKey: kCIInputImageKey)
return localFilter.?outputImage
}
}
Once we have 2~3 classes, we may find a way to DRY it up a bit.
| let image = CIImage(data: data)! | ||
|
|
||
| // found a way to get to CIImage from web resource -- can use this in a refactor later | ||
| // let webImage = CIImage(contentsOf: URL(string: aang)!) |
See: #4
As we learn how to Swift; we're stepping away from the SwiftUI/etc.
system and focusing more on building a library for doing the filtering.
We'll start with downloading a file from the Internet, saving it, then
running it through a LensStack, and saving a copy at each stage.
This way, we can start to diagnose if/how the filters are working
without futzing about with SwiftUI; which adds a confounding factor when
trying to figure things out.
TODO
name-original.pngname-noop.png