We have been developing smartphone applications for over 10 years, and recently we have been providing AI/deep learning based solutions that can be applied to web services, API services, and of course mobile app services as well.
PlayingBox is a cross-platform Dropbox video player.
Video screen: segment repeat, likes, and progress tracking.
List screen: custom playlists and quick organization.
Settings screen: flexible playback and app options.
Features:
- Dropbox video streaming
- Continue watching sync
- A-B segment repeat
- Custom playlists and favorites
- Playback speed and skip controls
- Multi-platform support (iOS/macOS/Android/Windows)
- Kotlin
- MVI (Clean Architecture)
- Jetpack Compose
- Kotlin Flows
- Orbit
- ContainerHost
- Dependency Injection
Development of an app where musicians compete,
and general users watch performances on YouTube,
vote, leave comments, and engage through the platform,
leading all the way to the final competition results.
- MVVM architecture using Retrofit2 and ViewModel
- Separation of UI, networking, and business logic
- Firebase messaging/crashlytics, glide
- Partial Kotlin adoption
- Custom UI with vector assets
- Material Design, Constraint Layout
- Integrated the open-source androidyoutubeplayer
A model trained with a 3D deep learning pipeline
resolves unnatural artifacts in face morphing.
The model can theoretically cover head rotations of about 90-180 degrees,
as well as vertical head movement.
It is trained by analyzing as many facial landmarks as possible
and vectorizing them into 3D representations.
Even when the angle changes, depth images make
natural face transitions possible.
Because the transformation is based on the user's own face,
seamless continuity is the key point.
The model achieved approximately 95% accuracy.
Depending on the scenario, we plan to upgrade it
by considering not only MobileNet,
but also models optimized for deepfake implementation.
Development of a video-based object recognition web/API service
by combining YOLOv5 and multiple open-source technologies.
For detected people, the service provides emotion, age, and gender analysis,
tagging of the same face (face clustering),
playback of only the segments where a specific face appears,
and frame-by-frame recognition results.