QwtSpectrogram, Change axis numbers

[Uwe]

Better remove all what you did so far and restart with something like that:

class SpectrogramData: public QwtRasterData
{
public:
    SpectrogramData()
    {
        m_intervals[ Qt::XAxis ] = QwtInterval( 0, 120 );
        m_intervals[ Qt::YAxis ] = QwtInterval( 0, 120 );
        m_intervals[ Qt::ZAxis ] = ...
    }
 
    virtual QwtInterval interval( Qt::Axis axis ) const
    {
        if ( axis >= 0 && axis <= 2 )
            return m_intervals[ axis ];
 
        return QwtInterval();
    }
 
    virtual double value( double x, double y ) const
    {
        double value = ...; // find the value corresponding to x/y in your vector
        return value;
    }
 
private:
    QwtInterval m_intervals[3];
};

class SpectrogramData: public QwtRasterData
{
public:
    SpectrogramData()
    {
        m_intervals[ Qt::XAxis ] = QwtInterval( 0, 120 );
        m_intervals[ Qt::YAxis ] = QwtInterval( 0, 120 );
        m_intervals[ Qt::ZAxis ] = ...
    }

    virtual QwtInterval interval( Qt::Axis axis ) const
    {
        if ( axis >= 0 && axis <= 2 )
            return m_intervals[ axis ];

        return QwtInterval();
    }

    virtual double value( double x, double y ) const
    {
        double value = ...; // find the value corresponding to x/y in your vector
        return value;
    }

private:
    QwtInterval m_intervals[3];
};

To copy to clipboard, switch view to plain text mode 

If you don't insist on using std::vector you could use QwtMatrixRasterData like demonstrated in the rasterview example. You get at least bilinear interpolation and a valid pixel hint for free.

Uwe

[kkaz]

So, as i am given to understand, i should map the values myself into the data, which i did and it seems to work!
I thought that just changing what values are displayed in the x-axis should be much easier.

The problem with the rasterview example is that if i use setValueMatrix() to set the data, without subclassing and using value(x,y) overloading, then the rendering time is much higher, i think. The subclassing way seems to be the fastest way.

Also, how many values are picked in the QwtInterval 0-120, and how are they picked. I mean, i want to achieve a certain resolution based on the data and, obviously, 2000*2000 cannot be rendered in a small window.

[Uwe]

I thought that just changing what values are displayed in the x-axis should be much easier.

You can set the axes manually using QwtPlot::setAxisScale or you rely on autoscaling, what is based on the bounding rectangle of the plot items. In your case the bounding rectangle depends on the intervals returned by raster data.
So there is nothing difficult about this - beside, that you are not setting the proper intervals. Return [0,120] and that's it.

The problem with the rasterview example is that if i use setValueMatrix() to set the data, without subclassing and using value(x,y) overloading, then the rendering time is much higher, i think.

You might not want to copy your data only because of API reasons, but there is no performance benefit from not using it.

Also, how many values are picked in the QwtInterval 0-120, and how are they picked.

The spectrogram has to create a QImage from the data, what is a matrix of RGB values. Position and size are found from the resolution of the data ( YourRasterData::pixelHint() ) and the resolution of the target device ( usually the screen ). Then it simply runs over the pixels of the image, calculates the corresponding position in plot coordinates and requests the value by calling YourRasterData::value(). Finally it maps the value into a RGB value and stores it in the image.

Calculating the value for a given position usually requires some sort resampling - this is what needs to be done inside of QwtRasterData::value(). Most implementations do some sort of nearest neighbor ( = simple and fast ), but of course you might lose significant pixels when the resolution of the image is too low. If this is not acceptable in your case you need to do calculate the value by taking adjacent values into count. In the end it is performance vs. quality.

When using QwtMatrixRasterData the resolution of the data can be calculated on its own ( you don't need to implement a pixelHint ) and bilinear interpolation as resampling algo is available. Or in other words: this is the option, for everyone who is not interested in these details.

Uwe