Hello~ I'm doing a object detection project and want to use Toyplot for visualization when training. The code I tried so far is:
import numpy as np
import toyplot
import toyplot.svg
from skimage.data import astronaut
img = astronaut()
w, h = img.shape[:2]
canvas = toyplot.Canvas(width=w, height=h)
mark = canvas.image(img, rect=(0, 0, w, h))
axes = canvas.cartesian(show=False, margin=0, padding=0)
bbox = np.float32([
[22, 33, 44, 55],
]) # x1, x2, y1, y2
mark = axes.rects(bbox[:, 0], bbox[:, 1], bbox[:, 2], bbox[:, 3], style={
'stroke': 'orange',
'stroke-width': 2,
'fill-opacity': 0.0
})
toyplot.svg.render(canvas, './vis.svg')However, the result seems incorrect. I expect a small rectangle but got a rectangle same size as the image. Are there any problem in the code? Or are there any other way to overlay bounding boxes on an image?
After tracing code, I found a solution. We need to specify xmin, xmax, ymin, ymax of cartesian to make cartesian.project work which is used by here. So
# Note the inversion of y-axis
axes = canvas.cartesian(show=False, margin=0, padding=0, xmin=0, xmax=w, ymin=h, ymax=0)gets the desired result!
@amoshyc - Toyplot makes a distinction between data marks and drawing marks. Data marks (like rectangles) have domain data, and require a set of axes (cartesian in this case) to map that data to canvas coordinates. Drawing marks (like images) don’t have a domain, they’re simply placed on the canvas. So in this case, you’re making your cartesian domain equal to the resolution of the image, so one domain unit = one pixel. As it happens, I already have a recent request to support rectangles as drawing marks, which would probably behave more like what you’re expecting. Cheers, Tim.
@tshead I see~ I' m looking forward to your future work. Currently what I need is to draw bounding boxes, keypoints, masks(polygons) on an image. An example is here. However, It seems that Toyplot doesn't have the ability to draw polygon and I would like to contribute. In this case, should I use Data Marks or Drawing Marks? If I stick to former, maybe I can utilize
I was using svgwrite previously, but it lacks of colormap and other high-level element. Are there any chance to make Toyplot able to add raw svg element directly? Although I don't think it is the purpose of this library. Thanks for the good library anyway XD
toyplot.graph but using latter is more logical.I was using svgwrite previously, but it lacks of colormap and other high-level element. Are there any chance to make Toyplot able to add raw svg element directly? Although I don't think it is the purpose of this library. Thanks for the good library anyway XD
@amoshyc - A data mark is most appropriate when you have a data domain (typically with units) that should be affected by your choice of projection (for example: a linear process doesn’t look linear when using a log projection). For your purposes, polygons as drawing marks are probably the semantically right approach - presumably, you want to draw polygons around features in your image, so image / canvas coordinates are the most logical way to think about it.
Also: using graph marks as a short turn workaround to implement polys is a clever idea, albeit semantically wrong - long-term, Toyplot tries to get the semantics of marks right, so that they can provide additional, mark-centric interaction.
Toyplot and Python 2
Gang:
Just a reminder that Python 2 will no longer be maintained after January 1, 2020 (https://pythonclock.org), and Toyplot is one of a large number of scientific projects that are ending support for Python 2 prior to that date (https://python3statement.org). So here’s what to expect over the next few weeks:
- Before December 31st, 2018, a Toyplot 0.18.0 release of the current master. This will be the final Toyplot release with Python 2 support.
- After January 1st, 2019, I will begin removing Python 2 portability-related code from master, and stop running Python 2 regression tests on Travis.
- Toyplot 0.18.0 will have its own branch (our releases are normally just tagged).
- Pull requests with bug fixes against the 0.18.0 branch will be accepted on a case-by-case basis, for an indeterminate amount of time.
FWIW, I’ve been doing virtually all of my Toyplot development on Python 3 for over a year, without complaint. You should consider doing the same with your work. It’s time. Don’t panic.
Cheers,
Tim
@amoshyc asked:
Are there any chance to make Toyplot able to add raw svg element directly?
Sorry I overlooked this. We don’t support embedding directly, because it would complicate support for non-SVG backends. However, you can do this pretty easily as a postprocessing step. See the documentation for
- https://toyplot.readthedocs.io/en/stable/toyplot.html.html#toyplot.html.render
- https://toyplot.readthedocs.io/en/stable/toyplot.svg.html#toyplot.svg.render
… in either case, you can get the output as a DOM tree, suitable for further manipulation. The structure of the DOM is pretty clean, and there are CSS classes you can use to anchor queries.
Toyplot 0.18.0 Released - highlights include improvements to the format module, and new support for ellipse and range visualizations. Most importantly, 0.18 will be the final Toyplot release with Python 2 support. As usual, all the details are at http://toyplot.readthedocs.io/en/stable/release-notes.html … it’s a Festivus miracle!
hannes is https://toyplot.readthedocs.io/en/stable/tick-locators.html#explicit-locators the "proper" way of labeling categories in a bar chart? my x axis is categorical
@hannes - yep, that’s the way to do it. While you’re at it, don’t forget that you can use rich text in your labels.
hannes i would like to render some bigger matrix
If you find that you’re doing it a lot, you can also do
import toyplot.config
toyplot.config.autoformat = “png”to make it the default (at the top of your notebook, say).
good day, I'm working on a little side project to visualize graph traversal using adjacency matrix multiplication using GraphBLAS. What I was hoping for was a tip on how to put two plots side by side in toyplot. For example, here's a static image of the layout i'm looking for https://github.com/michelp/pggraphblas/raw/master/docs/AdjacencyMatrixBFS.svg?sanitize=true
The static image shows one "step" in a matrix x vector multiplication using a boolean semiring. In addition to discovering how to do that layout with toyplot (which looks very amazingly cool btw!) if anyone has a tip on how i can even possibly animate the process i do eventually hope to many that possible, but first things first, being able to layout a graph next to its adjacency matrix is my first step
I have the code that generates the graph and the matrix, i'm just not sure how to put them next to each other in the same cell (i'm using jupyter)
I'm moving along pretty nicely with my project, but have run into one issue, is it possible to dynamically assign edge labels to a graph? On the examples at https://toyplot.readthedocs.io/en/stable/graph-visualization.html#edge-rendering all the edge labels are the same static value, but i'd like to assign my labels per edge to correspond to the value in the equivalent adjacency matrix cell.
hannes i often want to use data from dictionaries with toyplot, eg .keys() as x and .items() as y axes. unfortunately this leads to "TypeError: float() argument must be a string or a number, not 'dict_keys'" from https://toyplot.readthedocs.io/en/stable/_modules/toyplot/require.html#scalar_array because numpy's conversion to an array does not like python's dict views or iterators
hannes i can wrap it into a list() call but that seems ugly
hannes no idea why numpy does not support those but i wonder if adding some logic to toyplot's scalar_array function would make sense for this usecase?
hannes ```
hannes Example of what I would love to use (in an explicite way):import toyplot
data = {1: 5, 2: 3, 3: 4}
canvas = toyplot.Canvas()
axes = canvas.cartesian()
mark = axes.plot(
data.keys(),
data.values(),
)hannes https://toyplot.readthedocs.io/en/stable/tutorial.html#scatterplots uses a shared x axis for a stacked plot, is there a way to make a "true" scatterplot of irregular data?
hannes i have unique x and y axes for each series
hannes just use the default until it does not do what you need
hannes posted an image: 
hannes hm, when using matplotlib, one can get an image in there from a script itself by calling plt.show()
hannes i guess that is a super magic command
hannes sad how bad the state of relaying between messaging services still is :(