Implementation:Tencent Ncnn Mat Pixel Android

Knowledge Sources	Tencent_Ncnn
Domains	Android, Computer Vision
Last Updated	2026-02-09 19:00 GMT

Overview

Provides Android-specific convenience methods for converting between Android Bitmap objects (accessed via JNI) and ncnn Mat tensors, enabling direct image input from Android's native bitmap format.

Description

This file (213 lines) implements five Mat static and instance methods guarded by NCNN_PLATFORM_API and __ANDROID_API__ >= 9. The methods bridge the gap between Android's android.graphics.Bitmap Java class (accessed through JNI) and ncnn's Mat tensor format.

Each method follows the same pattern:

Retrieve bitmap info via AndroidBitmap_getInfo() to determine format and dimensions
Determine the source pixel format (PIXEL_GRAY for ANDROID_BITMAP_FORMAT_A_8 or PIXEL_RGBA for ANDROID_BITMAP_FORMAT_RGBA_8888)
Lock the pixel buffer with AndroidBitmap_lockPixels()
Delegate to the appropriate Mat::from_pixels* or to_pixels_resize method for format conversion
Unlock with AndroidBitmap_unlockPixels()

The five methods are:

Mat::from_android_bitmap() -- Basic conversion from bitmap to Mat with optional pixel format conversion.
Mat::from_android_bitmap_resize() -- Converts and resizes to target dimensions in one step.
Mat::from_android_bitmap_roi() -- Extracts a region of interest (ROI) from the bitmap.
Mat::from_android_bitmap_roi_resize() -- Extracts ROI and resizes to target dimensions.
Mat::to_android_bitmap() -- Writes Mat data back into an Android Bitmap.

All methods handle automatic pixel format negotiation using PIXEL_CONVERT_MASK and PIXEL_CONVERT_SHIFT bit manipulation to determine whether a format conversion is needed between the bitmap's native format and the requested target format.

Usage

Use these methods in Android JNI native code to load camera frames or gallery images directly into ncnn Mat tensors for inference, and to write results back to bitmaps for display. This eliminates the need for manual pixel buffer management in Android applications.

Code Reference

Source Location

Repository: Tencent_Ncnn
File: src/mat_pixel_android.cpp

Signature

// Convert Android Bitmap to ncnn Mat
static Mat Mat::from_android_bitmap(JNIEnv* env, jobject bitmap,
    int type_to, Allocator* allocator);

// Convert and resize in one step
static Mat Mat::from_android_bitmap_resize(JNIEnv* env, jobject bitmap,
    int type_to, int target_width, int target_height, Allocator* allocator);

// Extract region of interest from bitmap
static Mat Mat::from_android_bitmap_roi(JNIEnv* env, jobject bitmap,
    int type_to, int roix, int roiy, int roiw, int roih, Allocator* allocator);

// Extract ROI and resize
static Mat Mat::from_android_bitmap_roi_resize(JNIEnv* env, jobject bitmap,
    int type_to, int roix, int roiy, int roiw, int roih,
    int target_width, int target_height, Allocator* allocator);

// Write Mat back to Android Bitmap
void Mat::to_android_bitmap(JNIEnv* env, jobject bitmap, int type_from) const;

Import

#include "mat.h"
#include <android/bitmap.h>
#include <jni.h>

I/O Contract

Inputs

Name	Type	Required	Description
env	JNIEnv*	Yes	JNI environment pointer from the Android runtime
bitmap	jobject	Yes	Android Bitmap Java object reference
type_to	int	Yes	Target pixel format (e.g., ncnn::Mat::PIXEL_BGR, PIXEL_RGB)
target_width	int	For resize variants	Target width for resizing
target_height	int	For resize variants	Target height for resizing
roix, roiy	int	For ROI variants	Top-left corner of the region of interest
roiw, roih	int	For ROI variants	Width and height of the region of interest
allocator	Allocator*	No	Custom memory allocator (can be nullptr)
type_from	int	For to_android_bitmap	Source pixel format of the Mat data

Outputs

Name	Type	Description
return (from_*)	ncnn::Mat	New Mat containing the converted pixel data; empty Mat on unsupported bitmap format
bitmap (to_android_bitmap)	jobject	The Android Bitmap is written to in-place with Mat pixel data

Usage Examples

Loading Android Camera Frame for Inference

// In JNI native method:
extern "C" JNIEXPORT void JNICALL
Java_com_example_MyActivity_detect(JNIEnv* env, jobject, jobject bitmap)
{
    // Convert RGBA bitmap to BGR ncnn Mat
    ncnn::Mat in = ncnn::Mat::from_android_bitmap(env, bitmap,
        ncnn::Mat::PIXEL_BGR);

    // Or convert and resize to model input size
    ncnn::Mat in_resized = ncnn::Mat::from_android_bitmap_resize(env, bitmap,
        ncnn::Mat::PIXEL_BGR, 224, 224);

    // Run inference...
}

Writing Results Back to Bitmap

// Write detection result visualization back to bitmap
ncnn::Mat result_pixels = /* processed image */;
result_pixels.to_android_bitmap(env, output_bitmap, ncnn::Mat::PIXEL_RGBA);

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment