SegmentationWorker class

A dedicated background isolate for selfie segmentation inference.

This worker runs TFLite inference in a background isolate to avoid blocking the main UI thread, while using TransferableTypedData for efficient zero-copy data transfer (unlike IsolateInterpreter which uses slow nested list serialization).

Performance

• No UI blocking: Inference runs entirely in background isolate
• Fast data transfer: ~1ms via TransferableTypedData vs ~60-80ms with IsolateInterpreter
• Overall: ~30-50ms per inference (same as direct invoke, but non-blocking)

Usage

final worker = SegmentationWorker();
await worker.initialize();

// Process frames without blocking UI
final mask = await worker.segment(imageBytes);
print('Foreground: ${mask.data.where((v) => v > 0.5).length} pixels');

worker.dispose();

Memory Management

Model memory usage depends on the selected model variant (see SegmentationModel). Call dispose when done to free resources.