Skip to content

1.0.0-beta.3

Compare
Choose a tag to compare
@charliesantos charliesantos released this 25 May 16:40
· 71 commits to master since this release

1.0.0-beta.3 (May 25, 2021)

Improvements

  • The VideoProcessors now use WebAssembly to run TensorFlow Lite for faster and more accurate person segmentation. You need to deploy the tflite model and binaries so the SDK can load them properly. Additionally, this improvement requires Chrome's WebAssembly SIMD support in order to achieve the best performance. WebAssembly SIMD can be turned on by visiting chrome://flags on versions 84 through 90. This will be enabled by default on Chrome 91+. You can also enable this on versions 84-90 for your users without turning on the flag by registering for a Chrome Origin Trial for your website.

  • The segmentation model has been changed from BodyPix to MLKit Selfie Segmentation to improve segmentation accuracy.