Skip to content

Face Landmarker has shorter detection range on iOS compared to Web #5866

Open
@rcanbaba

Description

@rcanbaba

Have I written custom code (as opposed to using a stock example script provided in MediaPipe)?
None

OS Platform and Distribution
iOS 18, Xcode 16.1, macOS Sequoia

MediaPipe Version
MediaPipeTasksVision 0.10.21

Task Name
Face Landmark Detection

Describe the issue
We have observed a significant difference in face detection range between iOS and Web implementations of MediaPipe Face Landmark Detection.

On iOS, face detection fails beyond ~1.5 meters from the camera.
On the Web implementation, detection works beyond 3 meters, even though both platforms use the same face_landmarker.task model.
This behavior suggests that iOS has a shorter detection range.

Expected Behavior
Face detection should work at similar distances on both Web and iOS platforms, assuming the same model and similar camera conditions.

Is there a resolution scaling issue affecting iOS?
Does iOS process input frames differently before passing them to the model?
Is there any known parameter/configuration to extend the detection range?

Metadata

Metadata

Assignees

Labels

os:macOSIssues on MacOSplatform:iosMediaPipe IOS issuestask:face landmarkerIssues related to Face Landmarker: Identify facial features for visual effects and avatars.type:performanceExecution Time and memory heap, stackoverflow and garbage collection relatedtype:supportGeneral questions

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions