ios-vision

Hand and body pose detection on iOS uses VNDetectHumanHandPoseRequest and VNDetectHumanBodyPoseRequest from the Vision framework. These APIs analyze live camera feeds or recorded video to track joint positions and movement, enabling gesture recognition and motion analysis workflows.

Vision provides lower-level APIs for object detection, face detection, and pose tracking across images and video. VisionKit offers higher-level, pre-built UI workflows. Choose Vision for custom pipelines and fine-grained control; VisionKit for rapid document and standard analysis tasks.

Subject segmentation and foreground lifting on iOS uses VNGenerateForegroundInstanceMaskRequest to create instance masks that separate subjects from backgrounds. This generates binary or multi-instance segmentation masks you can apply to images or video frames for compositing and analysis.

Yes, Vision framework APIs work with live camera feeds, recorded media, and static images. Coordinate conversion and performance optimization are critical for real-time processing; batch requests and frame skipping strategies help manage latency on device.

iOS Vision requires coordinate conversion between image space, view space, and device orientation. Proper mapping ensures detections align with UI elements and camera previews, especially for pose tracking and segmentation masks in live-camera workflows.

Vision object detection works well for common subjects but may struggle with extreme angles, occlusion, or poor lighting. Face detection has accuracy trade-offs with performance; pose detection requires sufficient body visibility. Test detection quality on your target scenarios before production deployment.