Skip to content

@ghop02 ghop02 released this Feb 15, 2019 · 79 commits to master since this release

Highlights

  • Added pose estimation feature. You can use the FritzVisionPoseModel to detect human figures in images and video (Documentation).
  • Added ability to download FritzVision models over the air (Documentation).
  • Standardized feature APIs across all FritzVision models (API Reference).
  • Added ability to download custom models by tags (Documentation).
  • Added user-configurable metadata to Custom Models. (Documentation).
  • Simplified dependencies. All feature models are now included in the FritzVision target.
  • Fixed many broken Objective-C interfaces.

Breaking changes

  • All VisionModelOptions objects no longer can be configured during initialization. You can configure model options after creating an options object:
-  let modelOptions = FritzVisionLabelModelOptions(threshold: 0.5, numResults: 10)
+  let modelOptions = FritzVisionLabelModelOptions()
+  modelOptions.threshold = 0.5
+  modelOptions.numResults = 10
  • All models now run with a default imageCropAndScaleOption of .scaleFit. You will have to make the following changes to keep the same behavior:
-  let modelOptions = FritzVisionLabelModelOptions()
+  let modelOptions = FritzVisionLabelModelOptions()
+  modelOptions.imageCropAndScaleOption = .centerCrop

This is especially important for the object detection model as the crop and scale option affects the placement of the bounding boxes.

-  let modelOptions = FritzVisionObjectModelOptions()
+  let modelOptions = FritzVisionObjectModelOptions()
+  modelOptions.imageCropAndScaleOption = .centerCrop
  • Changed access to label and confidence on FritzVisionObject to simplify access:
- visionObject.label.label
- visionObject.label.confidence
+ visionObject.label
+ visionObject.confidence
Assets 2
You can’t perform that action at this time.