google-ai-edge/mediapipe v1.2.3

MediaPipe v1.2.3

on GitHub

Bazel changes

• Migrate mediapipe::NoDestructor to absl::NoDestructor (absl/base/no_destructor.h) in mediapipe/framework and remove obsolete mediapipe/framework/deps/no_destructor.h.
• Respect TensorBufferRequirements alignment in InferenceRunnerLiteRt
• Resolve file-path media parts in SemanticRetriever::Retrieve before embedding.
• [MediaPipe] Fix tensor view destruction deadlock on LiteRT run error.
• Register only the CPU accelerator in LiteRT environments for CPU-only runners.
• Add NPU dispatch library directory to C MpBaseOptions.
• Forward BaseOptions app telemetry in C++ tasks and Swift wrappers.
• Unify util/endian.h using absl/numeric/bits.h and remove deps/endian.h
• Format C++ code and fix includes in framework
• Format C++ code and fix includes in gpu
• Bump MediaPipe version to 1.2.3
• Support string input tensors in the MediaPipe LiteRT inference runner.
• Update Swift Package Manager files for the 1.2.3 iOS release.
• Added C API classifier options conversion to MPPClassifierOptions+Helpers

MediaPipe Tasks update

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

Android

• Unify MediaPipe Tasks option, method, and result names across Python, Web, Android, and iOS while preserving deprecated compatibility aliases, and update DevSite guides and code snippets.

iOS

• Add pure Swift microphone audio recording for MediaPipe Tasks iOS.
• Add pure Swift vision running mode, image processing options, and segmentation mask for MediaPipe Tasks iOS.
• Add pure Swift vision image input container for MediaPipe Tasks iOS.
• Add C API image conversion and ScopedMpImage to MPPImage+Utils.
• Add C API task runner helpers and VisionCallbackPool to MPPVisionTaskRunner.
• Remove legacy iOS C++ task runners, packet creators, and protobuf helpers.
• Migrate TextClassifier to MediaPipe Tasks C API.
• Migrate TextEmbedder to MediaPipe Tasks C API.
• Migrate LanguageDetector to MediaPipe Tasks C API.
• Migrate TextSummarizer to MediaPipe Tasks C API.
• Migrate TextProofreader to MediaPipe Tasks C API.
• Migrate AudioClassifier to MediaPipe Tasks C API.
• Migrate UniversalEmbedder to MediaPipe Tasks C API.
• Migrate ImageClassifier to MediaPipe Tasks C API.
• Migrate ImageEmbedder to MediaPipe Tasks C API.
• Migrate FaceDetector to MediaPipe Tasks C API.
• Migrate ObjectDetector to MediaPipe Tasks C API.
• Migrate FaceLandmarker to MediaPipe Tasks C API.
• Migrate HandLandmarker to MediaPipe Tasks C API.
• Migrate PoseLandmarker to MediaPipe Tasks C API.
• Migrate HolisticLandmarker to MediaPipe Tasks C API.
• Migrate GestureRecognizer to MediaPipe Tasks C API.
• Migrate ImageSegmenter to MediaPipe Tasks C API.
• Migrate InteractiveSegmenter to MediaPipe Tasks C API.
• Migrate InteractiveSegmenterLegacy to MediaPipe Tasks C API.
• Unify MediaPipe Tasks option, method, and result names across Python, Web, Android, and iOS while preserving deprecated compatibility aliases, and update DevSite guides and code snippets.

Javascript

• Remove on-disk cache options from the web audio LiteRT-LM tasks.
• Fix DecisionMaker Web SDK APIs, boolean scoring, and native in-memory CPU loading.
• Support incremental TextToSpeech stream input on the web.

Python

• Expose BooleanQuestion and ChoiceQuestion examples in the Decision Maker C API and Python.

MediaPipe Dependencies

• Explicitly include direct headers and BUILD dependencies.
• Update MediaPipe Tasks Web WASM files (CL 1019210657).
• Update MediaPipe Tasks Web WASM files (CL 1019312591).