You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
iOS: document and face auto-capture now run on a leaner TFLite engine instead of MediaPipe - about 12 MB smaller to download and roughly 8.5 MB smaller inside the app, detection behavior verified frame-for-frame identical on a physical-device matrix before release, and document inference about 40% faster.
iOS: auto-capture inference degrades gracefully through a runtime retry ladder; total engine failure still falls back to manual capture, never a crash.
Android: the native SDK gained an optional didit-sdk-autodetection-play artifact that runs auto-capture on the Google Play services ML runtime (~43 MB smaller universal APK). This package's diditSdkAndroidVariant options are unchanged in this release; wrapper-level opt-in support is planned as a follow-up.
Both platforms: new session diagnostics (engine selection and inference latency) help support teams pinpoint device-specific capture issues.