Java Views.extendMirrorSingle方法代码示例

import net.imglib2.view.Views; //导入方法依赖的package包/类
/**
 * Gaussian Smooth of the input image using intermediate float format.
 * @param <T>
 * @param img
 * @param sigma
 * @return
 */
public static <T extends RealType<T> & NativeType<T>> RandomAccessibleInterval<T> gaussianSmooth(
		RandomAccessibleInterval<T> img, double[] sigma) {
	Interval interval = Views.iterable(img);

	ImgFactory<T> outputFactory = new ArrayImgFactory<T>();
	final long[] dim = new long[ img.numDimensions() ];
	img.dimensions(dim);
	RandomAccessibleInterval<T> output = outputFactory.create( dim,
			img.randomAccess().get().createVariable() );

	final long[] pos = new long[ img.numDimensions() ];
	Arrays.fill(pos, 0);
	Localizable origin = new Point(pos);

	ImgFactory<FloatType> tempFactory = new ArrayImgFactory<FloatType>();
	RandomAccessible<T> input = Views.extendMirrorSingle(img);
	Gauss.inFloat(sigma, input, interval, output, origin, tempFactory);

	return output;
}
 

import net.imglib2.view.Views; //导入方法依赖的package包/类
private void iterateWithRandomAccessiblePartial( final Interval interval ) {
	final ExtendedRandomAccessibleInterval<T, Img<T>> extended = Views.extendMirrorSingle(input);
	final Cursor<FloatType> oc = Views.iterable(Views.interval(output, interval)).localizingCursor();

	final int n = input.numDimensions();
	final long[] min = new long[ n ];
	final long[] max = new long[ n ];
	interval.min( min );
	interval.max( max );
	for ( int d = 0; d < n; ++d )
	{
		min[d] -= 1;
		max[d] += 1;
	}
	final Interval raInterval = new FinalInterval( min, max );
	final RandomAccess<T> ra = extended.randomAccess( raInterval );
	float I, In, Ine, Ie, Ise, Is, Isw, Iw, Inw;

	while(oc.hasNext()) {

		oc.fwd();

		ra.setPosition(oc);

		I = ra.get().getRealFloat();
		ra.bck(1);
		In = ra.get().getRealFloat();
		ra.fwd(0);
		Ine = ra.get().getRealFloat();
		ra.fwd(1);
		Ie = ra.get().getRealFloat();
		ra.fwd(1);
		Ise = ra.get().getRealFloat();
		ra.bck(0);
		Is = ra.get().getRealFloat();
		ra.bck(0);
		Isw = ra.get().getRealFloat();
		ra.bck(1);
		Iw = ra.get().getRealFloat();
		ra.bck(1);
		Inw = ra.get().getRealFloat();

		oc.get().set( I - 1/8f * (In+Ine+Ie+Ise+Is+Isw+Iw+Inw));
	}

}
 

import net.imglib2.view.Views; //导入方法依赖的package包/类
@Override
public ExtendedRandomAccessibleInterval<T, F> calculate(F input) {
	return Views.extendMirrorSingle(input);
}
 

import net.imglib2.view.Views; //导入方法依赖的package包/类
@Override
public boolean process()
{
	// do not operate at the edge, 80% of the memory is a good idea I think
	final long memAvail = Math.round( cudaDevice.getFreeDeviceMemory() * ( percentGPUMem / 100.0 ) );
	final long imgBytes = numPixels() * 4 * 2; // float, two images on the card at once

	final long[] numBlocksDim = net.imglib2.util.Util.int2long( computeNumBlocksDim( memAvail, imgBytes, percentGPUMem, img.numDimensions(), "CUDA-Device " + cudaDevice.getDeviceId() ) );
	final BlockGenerator< Block > generator;

	if ( accurate )
		generator = new BlockGeneratorVariableSizePrecise( numBlocksDim );
	else
		generator = new BlockGeneratorVariableSizeSimple( numBlocksDim );

	final Block[] blocks = generator.divideIntoBlocks( getImgSize( img ), getKernelSize( sigma ) );

	if ( !accurate && blocks.length == 1 && ArrayImg.class.isInstance( img ) )
	{
		IOFunctions.println( "Conovlving image as one single block." );
		long time = System.currentTimeMillis();

		// copy the only directly into the result
		blocks[ 0 ].copyBlock( img, result );
		long copy = System.currentTimeMillis();
		IOFunctions.println( "Copying data took " + ( copy - time ) + "ms" );

		// convolve
		final float[] resultF = ((FloatArray)((ArrayImg< net.imglib2.type.numeric.real.FloatType, ? > )result).update( null ) ).getCurrentStorageArray();
		cudaconvolve.gauss( resultF, getImgSizeInt( result ), sigma, OutOfBounds.EXTEND_BORDER_PIXELS, 0 );
		IOFunctions.println( "Convolution took " + ( System.currentTimeMillis() - copy ) + "ms using device=" + cudaDevice.getDeviceName() + " (id=" + cudaDevice.getDeviceId() + ")" );

		// no copy back required
	}
	else
	{
		final RandomAccessible< net.imglib2.type.numeric.real.FloatType > input;
		
		if ( accurate )
			input = Views.extendMirrorSingle( img );
		else
			input = img;
		
		for( final Block block : blocks )
		{
			//long time = System.currentTimeMillis();
			final ArrayImg< net.imglib2.type.numeric.real.FloatType, FloatArray > imgBlock = ArrayImgs.floats( block.getBlockSize() );

			// copy the block
			block.copyBlock( input, imgBlock );
			//long copy = System.currentTimeMillis();
			//IOFunctions.println( "Copying block took " + ( copy - time ) + "ms" );

			// convolve
			final float[] imgBlockF = ((FloatArray)((ArrayImg< net.imglib2.type.numeric.real.FloatType, ? > )imgBlock).update( null ) ).getCurrentStorageArray();
			cudaconvolve.gauss( imgBlockF, getImgSizeInt( imgBlock ), sigma, OutOfBounds.EXTEND_BORDER_PIXELS, 0 );
			//long convolve = System.currentTimeMillis();
			//IOFunctions.println( "Convolution took " + ( convolve - copy ) + "ms using device=" + cudaDevice.getDeviceName() + " (id=" + cudaDevice.getDeviceId() + ")" );

			// no copy back required
			block.pasteBlock( result, imgBlock );
			//IOFunctions.println( "Pasting block took " + ( System.currentTimeMillis() - convolve ) + "ms" );
		}
	}

	return true;
}