Github user MrBago commented on a diff in the pull request:
https://github.com/apache/spark/pull/20168#discussion_r160267553
--- Diff: mllib/src/main/scala/org/apache/spark/ml/image/ImageSchema.scala
---
@@ -143,12 +174,12 @@ object ImageSchema {
val height = img.getHeight
val width = img.getWidth
- val (nChannels, mode) = if (isGray) {
- (1, ocvTypes("CV_8UC1"))
+ val (nChannels, mode: Int) = if (isGray) {
--- End diff --
In the code bellow we always call `toBytes` per channel so everything will
be cast to a 1 byte per channel image type. I think this is fine for standard
image types (ie, jpg, png, and the like).
Technically you can register image readers in java and use a BufferedImage
of `TYPE_CUSTOM` for more complex images, but in practice I think we'll want to
take a very different code path for these complex images so we'll want to
introduce a new method if we add support to handle them.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
