DropBG

KO | EN

gitlite — search

#json #rag #automation #golang #cpp #list #large-language-models #agents #tensorflow #database

(★ 9)

No description available.

File Explorer

ZIP Download

- - release.yml
- .gitignore
- AndroidProjectSystem.xml
- compiler.xml
- deploymentTargetSelector.xml
- gradle.xml
- markdown.xml
- migrations.xml
- misc.xml
- runConfigurations.xml
- build.gradle.kts
- proguard-rules.pro
- after.png
- before.png
- index.html
- logo.png
- - gradle-wrapper.jar
  - gradle-wrapper.properties
- libs.versions.toml
- after.png
- before.png
- screen_home_history.png
- screen_idle.png
- screen_result.png
.gitattributes
.gitignore
build.gradle.kts
gradle.properties
gradlew
gradlew.bat
README.md
settings.gradle.kts

# Use via CDN

jsDelivr

jsDelivr serves any public GitHub repository as a CDN with zero setup. Pick a version and a file to get a ready-to-paste link and snippet.

Version

File

Link

Example

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

<h1>DropBG</h1>

> **Remove image backgrounds on-device. 100% private. Zero network. Instant.**

[![Platform](https://img.shields.io/badge/Android-8%2B-3DDC84?logo=android&logoColor=white)](https://developer.android.com/about/versions/26)
[![Model](https://img.shields.io/badge/Model-RMBG%201.4-FF6B6B?logo=huggingface&logoColor=white)](https://huggingface.co/briaai/RMBG-1.4)
[![Engine](https://img.shields.io/badge/Engine-ONNX%20Runtime-005C97?logo=onnx&logoColor=white)](https://onnxruntime.ai)
[![Kotlin](https://img.shields.io/badge/Kotlin-2.3.20-7F52FF?logo=kotlin&logoColor=white)](https://kotlinlang.org)
[![Compose](https://img.shields.io/badge/Compose-BOM%202026.03-4285F4?logo=jetpackcompose&logoColor=white)](https://developer.android.com/jetpack/compose)
[![AGP](https://img.shields.io/badge/AGP-9.0.1-3DDC84?logo=android&logoColor=white)](https://developer.android.com/build/releases/gradle-plugin)
[![MinSDK](https://img.shields.io/badge/minSdk-26-brightgreen?logo=android&logoColor=white)](https://developer.android.com/about/versions/26)
[![TargetSDK](https://img.shields.io/badge/targetSdk-36-brightgreen?logo=android&logoColor=white)](https://developer.android.com/about/versions/14)

</div>

---

## Before / After

---

## Why DropBG?

Every background removal tool today either sends your images to a cloud server or requires a subscription. **DropBG** changes that — the entire RMBG-1.4 model runs on your device using ONNX Runtime with XNNPACK acceleration.

**Your images never leave your phone.**

---

## Security Posture

| Threat | Mitigation |
|--------|-----------|
| Image upload to cloud | **No upload**. All inference is on-device. |
| Data leakage | No internet permission required for core function |
| Privacy violations | Images processed entirely in local memory |
| Third-party tracking | Zero analytics SDKs, zero telemetry |

---

## Performance

Measured on mid-range Android (arm64-v8a):

| Metric | Value |
|--------|-------|
| Model load time | ~1–3s (cold start) |
| Inference (1024×1024) | ~1–2s (XNNPACK CPU) |
| Model size (quantized ONNX) | 42 MB |
| APK size | 88 MB (debug) |
| App + native libs overhead | ~46 MB |
| minSdk | API 26 (Android 8.0) |

---

## Architecture

```
┌──────────────────────────────────────────────────────┐
│                     DropBG                              │
│  ┌───────────────────────────────────────────────┐   │
│  │  Presentation Layer (Compose + Material 3)     │   │
│  │  ┌──────────────┐  ┌──────────┐  ┌─────────┐  │   │
│  │  │  MainScreen  │  │  Preview │  │  Stats  │  │   │
│  │  └──────┬───────┘  └──────────┘  └─────────┘  │   │
│  └─────────┼─────────────────────────────────────┘   │
│            │ collectAsStateWithLifecycle()              │
│  ┌─────────▼─────────────────────────────────────┐   │
│  │  ViewModel (MainViewModel)                     │   │
│  │  ┌──────────┐ ┌────────┐ ┌─────────────────┐  │   │
│  │  │ select   │ │process │ │ save / share    │  │   │
│  │  │ image    │ │ bg rm  │ │ result          │  │   │
│  │  └────┬─────┘ └───┬────┘ └─────────────────┘  │   │
│  └───────┼───────────┼──────────────────────────┘   │
│          │           │ suspend                          │
│  ┌───────▼───────────▼──────────────────────────┐   │
│  │  BackgroundRemover (Inference Engine)         │   │
│  │  ┌──────────────────────────────────────┐    │   │
│  │  │  ONNX Runtime + XNNPACK (CPU)       │    │   │
│  │  │  RMBG-1.4 (Quantized, 42 MB)        │    │   │
│  │  │  Preprocess → Inference → Postproc  │    │   │
│  │  └──────────────────────────────────────┘    │   │
│  └──────────────────────────────────────────────┘   │
└──────────────────────────────────────────────────────┘
```

### Key Design Decisions

- **MVVM**: Clean separation — Compose UI → ViewModel (StateFlow) → BackgroundRemover → ONNX Runtime
- **No DI framework**: Manual dependency injection via `AndroidViewModel` — keeps dependencies traceable
- **Single Activity**: `MainActivity` with edge-to-edge rendering
- **State-driven UI**: `sealed class UiState` drives all UI transitions (Idle → ImageSelected → Processing → Done / Error)
- **Before/After toggle**: Instant switch between original and processed result
- **Save & Share**: Built-in gallery save and system share sheet

---

## Tech Stack

| Category | Technology | Version |
|----------|-----------|---------|
| Language | Kotlin | 2.3.20 |
| Build System | Gradle | 9.1.0 |
| Android Plugin | AGP | 9.0.1 |
| UI Framework | Jetpack Compose (BOM) | 2026.03.01 |
| Design System | Material 3 | via BOM |
| Architecture | MVVM + StateFlow | — |
| ML Engine | ONNX Runtime Android | 1.21.0 |
| ML Backend | XNNPACK (CPU) | — |
| Inference Model | RMBG-1.4 (ONNX quantized) | 42 MB |

---

## Installation

| Method | Command |
|--------|---------|
| **Download APK** | Grab latest from Releases |
| **Install via ADB** | `adb install app-debug.apk` |
| **Build from source** | `./gradlew assembleDebug` |

> **Requirements:** Android 8.0+ (API 26), ~200MB free storage

---

## Quick Start

```bash
git clone https://github.com/rajumark/DropBG.git
cd DropBG

# Build debug APK
./gradlew assembleDebug

# Install on connected device
adb install app/build/outputs/apk/debug/app-debug.apk
```

---

## Project Structure

```
app/
├── src/main/
│   ├── assets/
│   │   └── rmbg14_quant.onnx           ← Quantized model (42 MB, LFS)
│   ├── java/raju/shingadiya/dropbg/
│   │   ├── MainActivity.kt              ← Entry point (edge-to-edge)
│   │   ├── inference/
│   │   │   └── BackgroundRemover.kt     ← ONNX Runtime engine
│   │   ├── viewmodel/
│   │   │   └── MainViewModel.kt         ← State & business logic
│   │   ├── ui/
│   │   │   └── MainScreen.kt           ← All Compose UI
│   │   └── theme/
│   │       ├── Color.kt                 ← Violet/Teal palette
│   │       ├── Theme.kt                 ← Material 3 theme
│   │       └── Type.kt                  ← Typography
│   ├── AndroidManifest.xml
│   └── res/
│       ├── values/
│       └── xml/
│           └── file_paths.xml           ← FileProvider config
├── build.gradle.kts
├── gradle/
│   └── libs.versions.toml               ← Version catalog
├── .github/workflows/release.yml        ← CI/CD
└── docs/index.html                       ← Landing page
```

**File count:** 7 Kotlin files, 1 YAML, 1 HTML.

---

## CI/CD

Every push to `main` with `#go` in the commit message triggers GitHub Actions to:

1. Checkout with LFS (model file)
2. JDK 17 setup (Temurin)
3. Build the debug APK (`assembleDebug`)
4. Extract version from `build.gradle.kts`
5. Create a GitHub Release with auto-generated body
6. Upload the APK as a release artifact

---

## Sizing

| Component | Size |
|-----------|------|
| RMBG-1.4 (quantized ONNX) | 42 MB |
| ONNX Runtime native libs | ~35 MB |
| Compose + Material 3 + Kotlin stdlib | ~8 MB |
| App code + resources | ~3 MB |
| **Total APK (debug)** | **~88 MB** |

---

## Model Details: RMBG-1.4

| Attribute | Value |
|-----------|-------|
| Architecture | Segformer (IS-Net derived) |
| Parameters | 44.1M |
| Input size | 1024×1024 |
| Quantization | INT8 (42 MB) |
| Framework | ONNX |
| Use case | Background removal (binary segmentation) |

> **License note:** RMBG-1.4 is source-available for non-commercial use. For commercial use, contact BRIA AI.

---

## License

MIT — See [LICENSE](LICENSE).

---

**Built for the Edge AI era**  
*Your images stay where they belong — on your device.*

</div>

DropBG

File Explorer

# Use via CDN

Was this content helpful?

Recent Feedback

Download README