Extending the image data viewer (dh5view/View3D)

The PYME image data viewer is implemented in PYME.DSView.dsviewer.DSViewFrame and can be invoked either by running dh5view from the command line, or by importing and calling PYME.DSView.View3D or PYME.DSView.ViewIm3D from within python code [1]. Most of the viewers image processing functionality is accomplished by plugins, and this guide will focus on writing new plugins.


dsviewer plugins, as described here, typically operate on a single image, require user interaction, and are not easily automated. If coding some image processing functionality which could be sensible to apply in an automated, or ‘batch’ mode, please consider Writing a Recipe Module instead, and then optionally wrapping a recipe module with a thin plugin.

Plugin Introduction

dsviewer plugins are python files which are located in the PYME.DSView.modules directory. Any file that is located in within that directory will be automatically detected and treated as a plugin.


PYME.DSView.modules and PYMEnf.DSView.modules [2] are currently the only locations where modules will be detected. A more flexible mechanism of module discovery is high on the TODO list.

Plugins must implement a function called Plug(dsviewer) which takes an instance of the current PYME.DSView.dsviewer.DSViewFrame, and can implement any additional python logic. It is good practice not to put too much processing logic in the plugin itself [3]. A typical Plug() method instantiates a class which stores plugin state, keeps a reference to the DSViewFrame object, and either adds menu items for it’s functions, or registers callbacks for overlays or GUI panels.

Adding menu items

Most simple plugins will want to add one or more menu items. DSViewFrame provides a helper function, DSViewFrame.AddMenuItem() to do this.

Wrapping a recipe module

The easiest, most flexible, and recommended way of writing a plugin is to wrap an existing or new recipe module. For modules derived from PYME.recipes.base.Filter (most modules which transform a single input image into a single output image) already have much of the necessary code implemented and the wrapping is simply as matter of registering the modules dsviewer_plugin_callback function to a menu item:

def Plug(dsviewer):
    from PYME.recipes.XXX import YYY
    dsviewer.AddMenuItem(menuName='Processing', itemName='Do Stuff', callback = lambda e : YYY.dsviewer_plugin_callback(dsviewer))

Somewhat more complicated wrappings can build on the template below (which more or less captures the internals of dsviewer_plugin_callback:

def DoStuff(image, parentWindow=None, glCanvas=None):
    from PYME.recipes.XXX import YYY
    from PYME.DSView import ViewIm3D

    mod = YYY(inputName='input', outputName='output')
    if mod.configure_traits(kind='modal'):
        namespace = {'input' : image}

        #Open the result in a new window. Setting the parent and glCanvas options keeps the UI happy
        #The glCanvas option enables the viewer to synchronize zooming and panning with the VisGUI point display if
        #both are running in the same process.
        ViewIm3D(namespace['output'], parent=parentWindow, glCanvas=glCanvas)

def Plug(dsviewer):
    dsviewer.AddMenuItem(menuName='Processing', itemName='Do Stuff', callback = lambda e : DoStuff(dsviewer.image, dsviewer, dsviewer.visgui))

More complex and legacy plugins

Very few existing plugins follow the above pattern (mostly for historical reasons), and although new plugins are encouraged to implement and wrap recipes, there will be times when this doesn’t quite fit. Anything which requires significant GUI interaction (e.g. annotation of images, creation of overlays, etc ... will most certainly not fit the recipe module pattern. As such here’s a brief outline of what is exposed and what you can do. To fully understand the options, some examination of the code and existing modules is likely to be needed.

A PYME.DSView.dsviewer.DSViewFrame instance exposes three important attributes:

  1. dsviewer.image : A reference to the currently displayed ImageStack object.
  2. dsviewer.do : A reference to a PYME.DSView.displayOptions.DisplayOpts instance which stores the display settings for the current image. This is useful for determining the current position in the stack, for extracting manual threshold levels, and for setting overlays.
  3. dsviewer.view : A reference to the current view class (not commonly used).

A quick overview of the GUI components that plugins can alter in addition to menus is given below.


An example of a panel and overlay generated by a plugin (in this case PYME.DSView.modules.flowView).


One way of adding GUI functionality is to add ‘panels’ to the left side of the image view.


The interface for adding panels is currently pretty atrocious, and requires you to know more about the inner workings of wxpython and DSViewFrame than you probably want to. Some of the worst aspects will hopefully be refactored out at some point in the future, but for now, here be dragons.

Adding panels is accomplished by registering a callback that is called on window creation (and when new modules are loaded) which generates the panel.

An example (abbreviated/adapted from PYME.DSView.modules.particleTracking) is given below:

import wx
from traits.api import HasTraits, Int, Bool

class StuffDoer(HasTraits):
    someProperty = Int(7)
    anotherProperty = Bool(False)

    def __init__(self, dsviewer):
        dsviewer.paneHooks.append(self.GenStuffPanel) # this registers this panel

    def GenStuffPanel(self, _pnl):
        # this function will be called on window creation and whenever the side panel is rebuilt (e.g. when a new module is loaded)
        item = afp.foldingPane(_pnl, -1, caption="Stuff Settings", pinned = True)

        pan = self.edit_traits(parent=item, kind='panel')

        bDoStuff = wx.Button(item, -1, 'Do Stuff')
        bDoStuff.Bind(wx.EVT_BUTTON, self.OnDoStuff)


def Plug(dsviewer):
    dsviewer.stuff_doer = StuffState(dsviewer)

When possible, using Traits (as illustrated above) to generate the actual GUI, rather than hand-coding an interface in wxpython will save a lot of pain.


The DSViewFrame overlay system is a low level interface which allows plugins to draw overlays on top of the currently displayed image. To use overlays you will have to get familiar with reasonably low level wxpython programming, specifically operations on wx.DC objects.

You need to:

  1. Define a function with a signature matching DrawOverlays(view, dc) (or optionally DrawOverlays(self, view, dc) if a class method).
  2. Register this function by appending it to the overlays list in the display options, e.g. dsviewer.do.overlays.append(DrawOverlays)

The two parameters are the view object (an instance of PYME.DSView.arrayViewPanel.ArrayViewPanel) and the wx.DC instance the overlay should be drawn to. The view object is mainly useful as it provides a number of functions for mapping pixel co-ordinates to screen co-ordinates and vice-versa.

These are:

  • _ScreenToAbsCoordinates()
  • _AbsToScreenCoordinates()
  • _PixelToScreenCoordinates()
  • _PixelToScreenCoordinates3D()
  • _drawBoxPixelCoords()


The overlays mechanism needs a lot of work, including but not limited to removing the leading underscores from the co-ordinate transformation functions (these were initially written for class internal use, but have proved very outside the class as well).

At the moment, all overlay functions get called and it is up to the individual plugin to decide whether to display or not. On the TODO list is to add central control of overlay toggling and to change to using a registration function rather than appending to a list.


dsviewer supports 4 selection modes: point point, box box, straight line line, and curve squiggle. Box and line selections are reasonably self explanatory. Both cases are specified by two points, which either represent diagonally opposed corners of a box or endpoints of a line. Curve selections are a freeform curve, stored as a series of (x,y) co-ordinates. Both line and curve selections can have a width.


expand this description.



The program should be running a wxpython event loop. This will always be the case if called within one of the PYME GUI programs (dh5view, VisGUI, PYMEAcquire). If you want to call View3D or ViewIm3D from an ipython session you will need to run ipython --gui=wx to make sure the wx event loop is running. In an ipython/jupyter notebook you will need to use the %gui wx magic before running View3D.

Running from an ipython notebook with anaconda on OSX requires some additional fiddling - you have to change the shebang of /PATH/TO/anaconda/bin/ipython to point to the framework copy of python (usually PATH/TO/anaconda/python.app/Contents/MacOS\python) so that ipython notebooks can access the display without dying.

[2]PYMEnf is a module which is used internally within the Baddeley and Soeller groups and contains code that we cannot distribute due to licensing restrictions, contains sensitive information, or for some other reason is not ready for public release.
[3]Although the model-view-controller pattern is poorly followed in the majority of PYME code, it is useful to think of plugins existing at the controller level - providing the interface between image processing routines and libraries and the view code. That said, a lot of existing plugin code includes both GUI and program logic.