Archive

Archive for the ‘GSOC2015’ Category

Matplotlib for Kivy Final Notes

August 26th, 2015 No comments

As outcome of one of the projects from the google summer of code 2015 two packages were created and are available as kivy garden packages.

These two packages can be used separately or can be combined as shown in https://andnovar.wordpress.com/2015/08/06/corner-detection-on-strokes-and-strokes-annotations-on-matplotlib/.

Warning:

– If you are using dash_list attribute when creating a line in a figure, it will not work until the update is done in the next kivy version.
– The backend only works with the kivy >= 1.9.1. Some changes were done in the Widget class.
– Matplotlib backend has not been tested on android.

Both packages have been tested and debugged, however it is highly encouraged to submit any issue found while using them. Some of its advantages and weaknesses are exposed in the following paragraphs:

The first package is garden.matplotlib which is a matplotlib backend for kivy, it allows to create applications using pyplot instructions or to embed figures into a kivy application. There are two possible backends that can be invoked:

– backend_kivy which does the render using kivy graphics instructions.
– backend_kivyagg which does the render using an static image texture.

Both of the backends can be connected to the default matplotlib events https://andnovar.wordpress.com/2015/06/15/connecting-events-between-kivy-and-matplotlib/. Additionally another widget called NavigationToolbar can be instantiated and used with both backends, a better description can be found in https://andnovar.wordpress.com/2015/08/06/navigation-toolbar-with-matplotlib-events-connected/. A FigureCanvas can be used without a NavigationToolbar but a NavigationToolbar needs a FigureCanvas to be instantiated. The main advantage of this package is the ability to be added as another Kivy widget inside a Kivy application through the use of matplotlib instructions.

There are some features that were not implemented and the users should be aware of:

– backend_kivy and backend_kivyagg are non interactive backends (http://matplotlib.org/faq/usage_faq.html#what-is-interactive-mode)
– gouraud shading in backend_kivy
– configure_subplots button in the NavigationToolbar is not implemented.

Known bugs:

– When zooming on an area with not enough graphics instructions the render is not properly done. https://github.com/andnovar/kivy/issues/72

In the same way there are some features implemented in backend_kivy that are not implemented in other graphics instructions based backends.

– clip path when rendering an image.
– draw_path_collection method for optimization.
– draw_markers for optimization.

We believe most of matplotlib current capabilities have been considered during this backend implementation. However, we encourage to submit any capability missed you may find.

Categories: GSOC2015, kivy, matplotlib Tags: , ,

N$ recognizer kivy implementation

August 18th, 2015 No comments
Categories: GSOC2015, kivy, python Tags: , , ,

Corner detection on Strokes and Strokes annotations on Matplotlib

August 6th, 2015 2 comments

Following my proposal for the Google Summer of Code 2015, an algorithm to detect corners on strokes was implemented and it is going to be polished and tested with a template matching classifier. This implementation follows Wolin approach in his paper: “ShortStraw: A Simple and Effective Corner Finder for Polylines” which can be downloaded from: http://faculty.cse.tamu.edu/hammond/publications/pdf/2008WolinSBIM.pdf

A brief explanation of the algorithm consists on sampling each one of the points in order to have equal distance between them. Then a window of +/-3 points is used and the average or mean of the distances is calculated. The distance between the initial point and the last point of the window form a straw. The straw is smaller in the cases that the window of points contain a corner.

corners strokesFrom the figure can be seen that pretty much corners are being detected. However a post processing still needs to be performed. In addition, a stroke canvas behavior was added to a figure canvas widget so annotations can be done on top of the graphs as can be seen in the figure below. The yellow background and the blue dots are just debugging information.

backend_kivy with annotations

backend_kivy and backend_kivyagg working with matplotlib 1.4.3, python2 and python3.

August 6th, 2015 No comments

Initially backend_kivy was implemented using matplotlib 1.5.1dev which is an unstable release. Small modifications were made based on the changes matplotlib has between 1.4.3 (stable) and 1.5.1dev. Some of the main points that were updated were the generation of MathText and the lines rendering in draw_path. Additionally in the case of backend_kivyagg a read-only memory view was sent from the agg renderer to the kivy backend, a writable copy had to be created due to an exception in cython for loading read-only buffers. Some other improvements involve a bug when the application was being closed which was fixed by adding an on_close event in the app and setting the global var app to None, another improvement is the generation of images from the plots. In the case of backend_kivy the figure canvas widget is being exported to png using the export_to_png(filename). A bug was reported when the FigureCanvasKivy is added to a BorderLayout and the window is resized. For instance if we have a button on top of the figure canvas, when resized the figure canvas is resized to the size of the window moving the button up but then is scaled down again and the dimensions are right. This arises in the backend_kivy due to a small delay that keeps the calculation of the positions in this case of the button out of the canvas. In backend_kivyagg this does not happen due to a fast transition so the button position is calculated correctly. (still an open issue and working on that for backend_kivy.)

Navigation Toolbar with matplotlib events connected

August 6th, 2015 No comments

NavigationToolbar is the last visual aspect from the backend that has been added with all the possible events connected to matplotlib. The main changes performed on the implementation involved to connect each one of the toolbar elements to the respective matplotlib calls. Additionally, the figure canvas is separated from the toolbar and both are children of a Float Layout inside an app generated via pyplot. The FigureCanvasKivy can be called independently of the NavigationToolbar but not in the opposite direction.

# updated version of the toolbar generation.
def _init_toolbar(self):
        basedir = os.path.join(rcParams['datapath'], 'images')
        actionview = ActionView()
        actionprevious = ActionPrevious(title="Navigation", with_previous=False)
        actionoverflow = ActionOverflow()
        actionview.add_widget(actionprevious)
        actionview.add_widget(actionoverflow)
        actionview.use_separator = True
        self.actionbar.add_widget(actionview)
        for text, tooltip_text, image_file, callback in self.toolitems:
            if text is None:
                actionview.add_widget(ActionSeparator())
                continue
            fname = os.path.join(basedir, image_file + '.png')
            if text in ['Pan', 'Zoom']:
                action_button = ActionToggleButton(text=text, icon=fname)
            else:
                action_button = ActionButton(text=text, icon=fname)
            action_button.bind(on_press=getattr(self, callback))
            actionview.add_widget(action_button)

Navigation Toolbar Kivy

For matplotlib users the interface can look very familiar. However, I consider a brief explanation is necessary on the function of each button:

Home: Resets the plot axes to the initial state.
Left: Undo an operation performed.
Right: Redo an operation performed.
Pan: Allows to drag the plot.
Zoom: Allows to define a rectangular area to zoom in.
Configure: Not implemented yet but, loads a pop up for repositioning elements.
Save: Loads a Save Dialog to generate an image.
Configure and Save will be added at some point the next week.

Categories: GSOC2015, kivy, matplotlib Tags: , ,

Improvements to the RendererKivy (Part 2)

July 22nd, 2015 No comments

The following modifications have been implemented in these days:

  • Fixed problem when rendering concave polygons. It can be seen from previous images that in the case of an arrow and a wedge there was a defect on the rendering due to the triangulation algorithm to fill it. The mesh renderer uses “triangle_fan” and since all the points form triangles always with the hub in concave polygons this was a problem. It was fixed by splitting concave polygons into convex ones. This was implemented using Tesselator a utility class in the graphics kivy package.
  • Path was being closed when it should not. It has been fixed that a path line was being closed by default, the initial point was removed so the line does not try to close a path when it is not necessary.
  • Elements out of the figure axes are not rendered. The problem here is that everything was being drawn in the same layer. A stencil widget is used as a mask so everything out of bounds it is hidden. In the second figure, 4 stencil instructions are created to show just the graphics instructions visible in the clip rectangle.
  • Focus behavior was added to the FigureCanvasKivy. Keyboard events are automatically bound by the focus behavior.
  • Added support for multiple dash length and offset in a same line. Line can receive a list with the values of length and offset for the respective rendering. For instance [10, 5, 20, 10] would create a line with 10 visible points then 5 hidden 20 visible and 10 hidden. This can be seen on the figure of the sin function below

dynamic dashes in a line4 stencil widgetsartist demointegral demo

Categories: GSOC2015, kivy, matplotlib Tags: , ,

Improvements to the RendererKivy

July 17th, 2015 No comments

The main improvements to the Renderer involved to run a test for the backend_kivy with as many as possible examples. From the test we figured out there were some implementations missed. The following are the main improvements done to the Renderer this week:

  • The rendering of the text is not blurry anymore. This was fixed by working with int values instead of float, it seems like opengl was trying to give some percentage of the pixel value to the previous and next pixel giving this effect.
  • Math text such as integrals, derivatives and so on can be added to the figure canvas. Matplotlib provides a very useful math text parser which receives the string and returns an object that can be embedded on the kivy canvas.
  • Images as textures can be added inside a Figure and outside as well. Matplotlib allows to generate some interpolation graphs which can be inserted into the axes and give some fancy visualization of data. Since color in kivy’s renderering is multiplicative, you would want the color to be white or the texture will be ‘tinted’ as it will be something other than 1.* color channel at pixel. Initially the problem was that the texture was invisible but it was solved by changing the Color to white as it is explained before. When a figure is not present the image is added directly to the canvas as a texture.

Some examples can be seen in the pictures below. Additionally there are two known problems, the first one is that the rendering is out of the axes and the second one is related to the path and mesh creation. This problems can be seen specifically in the integral example and the electric dipole with the arrows and magnetic fields.

gradient bar imshow without figure

Read more…

Categories: GSOC2015, kivy, matplotlib Tags: , ,

Navigation Toolbar for #kivympl using an Action Bar

July 17th, 2015 No comments

After my last post I started the implementation of a Navigation Toolbar for the backend. For the Navigation Toolbar for kivy we decided to experiment with an ActionBar from the set of kivy widgets. It is a very early version and not implemented at all, but this is a snippet of how it would be the layout and the result can be seen on the image below. The only element being used for now are the icons for the action buttons so we can have an idea of how it would look like.  The behavior of each one of the items will be defined on the callbacks. The NavigationToolbar2Kivy extends from NavigationToolbar2 which is the base class.

def _init_toolbar(self):
        basedir = os.path.join(rcParams['datapath'], 'images')
        actionbar = ActionBar(pos_hint={'bottom': 1.0})
        actionview = ActionView()
        actionbar.add_widget(actionview)
        for text, tooltip_text, image_file, callback in self.toolitems:
            if text is None:
                # insert a separator
                continue
            fname = os.path.join(basedir, image_file + '.png')
            action_button = ActionButton(text=text, icon=fname)
            actionview.add_widget(action_button)
        self.canvas.add_widget(actionbar, canvas='after')

Navigation Toolbar Kivy MPL

Kivy matplotlib backend with right text positioning and rendering.

July 8th, 2015 No comments

In order to fix the problem of the layout of the text inside the canvas, an important function had to be overwritten with the information of the bounding box for each text inside it. The function is get_text_width_height_descent in this, one calculates the width and height and return them. Inside matplotlib these values are important to calculate the positioning of the elements according to the layout set by the user. Additionally, in the draw_text method transformations and styling is applied to the text. At the end we can have a working backend with different fonts and styles that are mapped between matplotlib and kivy capabilities.

Another inconvenient in the rendering of the graphs were some glitches while drawing. I was calculating u and v to create the mesh for the polygons when this was not really necessary. After some debugging and analysis of the vertices the final implementation generate the meshes and lines for all the elements in the canvas as follow:

Screenshot from 2015-07-07 12:18:28 example mpl kivy

As can be seen, now it is getting closer to how the static image implementation looks like from the previous post.

Kivy backend using Line and Mesh graphics instructions

June 26th, 2015 2 comments

This is the first prototype for the backend, points are extracted from the path an transformed into polygons or lines. Line and Mesh are used in the kivy side to render these objects in a widget canvas. Labels are used for displaying text. Below some examples can be found. The lines are not well defined and the next is to optimize this drawing as well as the text. Some attributes should be added to the labels, positioning is another problem.

matplotlib examples kivy matplotlib kivy matplotlib kivy matplotlib kivy

Read more…