github google-ai-edge/mediapipe v0.10.33
MediaPipe v0.10.33

13 hours ago

Build changes

  • [mediapipe] mark mediapipe/gpu:gl_context incompatible on Windows
  • [mediapipe] automatically disable gpu (OpenGL) when building for Windows
  • [mediapipe] update visibility of gpu:disable_gpu_flag

Framework and core calculator improvements

  • Add Nearest Neighbor interpolation to WarpAffineCalculator.
  • Add a Tensor DebugString() function that formats the tensor numpy-style
  • Add a Tensor test util
  • Create a C API for Holistic Landmarker
  • Remove the CPU-only MediaPipe LLM Inference Engine
  • Add an AddMultiStreamCallback version that takes a packet map
  • BUILD rule cleanup
  • API3 subgraph support
  • Adds LogImage for mediapipe::Image
  • Add support for FULL_RANGE face detection task
  • Ads optional RESET input stream to SpectrogramCalculator
  • Simplify TaskRunner initialization by adding an options object
  • Plumb the host platform and version for all MP Tasks users.
  • MediaPipe: Add ROI validation in ImageToTensorOpenCvConverter

MediaPipe Tasks update

This section should highlight the changes that are done specifically for any platform and don't propagate to
other platforms.

Javascript

  • Add tests for full-range face detector model.

Python

  • Re-adding old sources for Holistic Landmarker
  • Re-add Holistic Landmarker to Python
  • Small patch in old LLM converter
  • Support int2 quantization serialization in MP converter
  • Removing temporary flag in MP LLM converter

Don't miss a new mediapipe release

NewReleases is sending notifications on new releases.