v1.2.1 — About & Social Links

SHARP QNN v1.2.1

English:

Patch release — About section and social links.

New Features

• About section in Settings with GitHub project and Bilibili links

Changes

• README badges now include Bilibili
• Version bumped to 1.2.1

Requirements

• Snapdragon 8 Gen 2+ with HTP v73+
• Android 12.0+ (ARM64)

Installation
Download app-release.apk below and install on your device.
Pre-converted DLC models can be downloaded in-app from the Models page.

---

中文：

补丁版本 — 关于页面与社交链接。

新功能

• 设置页面新增"关于"选项卡，含 GitHub 项目与 B站链接

变更

• README 徽章新增 B站
• 版本号升至 1.2.1

系统要求

• 骁龙 8 Gen 2+，HTP v73+
• Android 12.0+（ARM64）

安装
下载下方 app-release.apk 安装即可。
预转换 DLC 模型可在应用内模型页面一键下载。

v1.2 — AnyCalib Focal Estimation & Model Hub

SHARP QNN v1.2

English:

Third release — AI-powered focal length estimation and enhanced model management.

New Features

• AI focal length suggestion powered by AnyCalib (ray-field estimation without EXIF)
• Editable focal length input with 2-decimal precision
• Extended models section with separate FP16 model download card
• One-click compile all uncompiled models with unified progress tracking
• Conditional download/import buttons (auto-disable when models present)

Improvements

• Image preprocessing uses center crop before resize for undistorted inference
• Diagnostic logs downgraded to debug level for cleaner release output
• README updated with AnyCalib project, paper, and Hugging Face repository references

Fixes

• Fixed status label always showing English when model not imported (Chinese locale)
• Fixed model loading failure by adopting stepwise model loading

Requirements

• Snapdragon 8 Gen 2+ with HTP v73+
• Android 12.0+ (ARM64)

Installation
Download app-release.apk below and install on your device.
Pre-converted DLC models can be downloaded in-app from the Models page.
The AnyCalib focal estimation model requires a separate download from the Extended Models section.

---

中文：

第三个版本 — AI 焦距估计与增强模型管理。

新功能

• 基于 AnyCalib 的 AI 焦距建议（无 EXIF 时通过射线场反推焦距）
• 可编辑焦距输入，支持两位小数精度
• 扩展模型栏目，独立 FP16 模型下载卡片
• 一键编译所有未编译模型，统一进度追踪
• 下载/导入按钮条件启用（模型已存在时自动禁用）

改进

• 图像预处理采用中心裁剪后缩放，避免推理变形
• 诊断日志降级为 debug 级别，Release 版本输出更干净
• README 新增 AnyCalib 项目、论文及 Hugging Face 仓库引用

修复

• 修复中文环境下未导入状态标签始终显示英文
• 修复模型加载失败问题，采用分步加载模型

系统要求

• 骁龙 8 Gen 2+，HTP v73+
• Android 12.0+（ARM64）

安装
下载下方 app-release.apk 安装即可。
预转换 DLC 模型可在应用内模型页面一键下载。
AnyCalib 焦距估计模型需在扩展模型栏目单独下载。

v1.1 — PLY Optimization & Performance

SHARP QNN v1.1

English:

Second release — PLY optimization, image picker, and performance improvements.

New Features

• PLY 3D Gaussian optimization via GaussSimplify (kNN Merge, Opacity Prune, SOR)
• System image picker with album bucket filtering

Improvements

• Parallelized model download for faster preparation
• Optimized image metadata reading
• Improved UI animations and interactions

Fixes

• Fixed various bugs

Changes

• License: MIT → GPL 3.0
• Supports Snapdragon 8 Gen 2+ (HTP v73+) only

Requirements

• Snapdragon 8 Gen 2+ with HTP v73+
• Android 12.0+ (ARM64)

Installation
Download app-release.apk below and install on your device.
Pre-converted DLC models can be downloaded in-app from the Models page.

---

中文：

第二个版本 — 新增 PLY 优化、图片选择器及性能提升。

新功能

• 基于 GaussSimplify 的 PLY 3D 高斯优化（kNN 合并、透明度剔除、SOR）
• 系统图片选择器，支持相册桶过滤

改进

• 模型下载并行化，准备阶段更快
• 图片元数据读取优化
• 优化 UI 动画与交互体验

修复

• 修复若干 Bug

变更

• 许可证：MIT → GPL 3.0
• 仅支持骁龙 8 Gen 2+（HTP v73+）

系统要求

• 骁龙 8 Gen 2+，HTP v73+
• Android 12.0+（ARM64）

安装
下载下方 app-release.apk 安装即可。
预转换 DLC 模型可在应用内模型页面一键下载。

v1.0.0 — Initial Release

SHARP QNN v1.0.0

First public release of ML-Sharp-QNN — single-image 3D Gaussian Splatting on Android via Qualcomm QNN HTP.

Features

• Fully offline inference on Snapdragon Hexagon DSP
• Single image → 3D Gaussian Splat (.ply)
• One-tap model download (HuggingFace / HF-Mirror)
• Bilingual UI (Chinese / English)
• EXIF-aware depth estimation

Requirements

• Snapdragon device with Hexagon DSP (SD 8 Gen 2+)
• Android 12.0+ (ARM64)

Installation

Download app-release.apk below and install on your device.
Pre-converted DLC models can be downloaded in-app from the Models page.