Introduction to dash-canvas
Note: dash-canvas is a legacy package. The recommended way to annotate images is to use the drawing tools of plotly figures.
dash-canvas is a module for image annotation and image processing using Dash. It provides both the DashCanvas object for drawing and annotations on images, and a set of utility functions to process images using the annotations.
dash-canvas can be used in various fields in which user interaction with images is required, such as quality control in industry, identification and segmentation of cells or organs in life and medical sciences, quantification of phases in materials and geosciences, construction of training sets for machine learning, etc.
Install dash-canvas with
pip install dash-canvas
The source is on GitHub at plotly/dash-canvas.
DashCanvas: a canvas object for annotations
Let’s get started with a simple canvas object.
from dash import Dash, html
from dash_canvas import DashCanvas
app = Dash(__name__)
app.config.suppress_callback_exceptions = True
app.layout = html.Div([
html.H5('Press down the left mouse button and draw inside the canvas'),
DashCanvas(id='canvas_101')
])
if __name__ == '__main__':
app.run(debug=True)
Press down the left mouse button and draw inside the canvas
You can draw inside the object with the freehand tool, and use the tool buttons to draw lines, zoom in and out, pan, select objects and move them inside the canvas.
DashCanvas comes with a set of properties which can be adjusted to control the geometry of the canvas, the default tool and its properties. You can pass a background image either as a filename (filename property) or as a data string (image_content property); more examples below.
from dash import Dash, html
from dash_canvas import DashCanvas
app = Dash(__name__)
filename = 'https://raw.githubusercontent.com/plotly/datasets/master/mitochondria.jpg'
canvas_width = 500
app.layout = html.Div([
DashCanvas(id='canvaas_image',
tool='line',
lineWidth=5,
lineColor='red',
filename=filename,
width=canvas_width)
])
if __name__ == '__main__':
app.run(debug=True)
The height of the canvas is adjusted automatically by keeping the aspect ratio of the background image.
Basic callbacks to modify DashCanvas properties
Like any Dash component, the properties of a DashCanvas can be modified by other components, via callbacks. Please be sure to have read about Basic Callbacks in the Dash Fundamentals.
from dash_canvas import DashCanvas
from dash import Dash, html, dcc, Input, Output, callback
import dash_daq as daq
filename = 'https://www.publicdomainpictures.net/pictures/60000/nahled/flower-outline-coloring-page.jpg'
canvas_width = 300
app = Dash(__name__)
app.layout = html.Div([
html.Div([
DashCanvas(
id='canvaas-color',
width=canvas_width,
filename=filename,
hide_buttons=['line', 'zoom', 'pan'],
)
], className="six columns"),
html.Div([
html.H6(children=['Brush width']),
dcc.Slider(
id='bg-width-slider',
min=2,
max=40,
step=1,
value=5
),
daq.ColorPicker(
id='color-picker',
label='Brush color',
value=dict(hex='#119DFF')
),
], className="three columns"),
])
@callback(Output('canvaas-color', 'lineColor'), Input('color-picker', 'value'))
def update_canvas_linecolor(value):
if isinstance(value, dict):
return value['hex']
else:
return value
@callback(Output('canvaas-color', 'lineWidth'), Input('bg-width-slider', 'value'))
def update_canvas_linewidth(value):
return value
if __name__ == '__main__':
app.run(debug=True)
Brush width
In the example above, a slider dcc.Slider and a color picker daq.ColorPicker are used to adjust the width and color of the drawing brush. We just created an image coloring tool in a few lines of code! You can learn more about available components in the
component libraries
section of the Dash documentation. Also note that the set of available buttons has been restricted through thehide_buttons properties, in order to keep the app design simple.
Retrieving the geometry of annotations and using utility functions
The geometry of annotations can be retrieved by pressing the bottom-right button of the DashCanvas. This button is called “Save” by default; the name can be customized through the goButtonTitle property. This button updates the json_data property of DashCanvas, which is a JSON string with information about the background image and the geometry of annotations.
from dash import Dash, html, dash_table, Input, Output, callback
from dash.exceptions import PreventUpdate
from dash_canvas import DashCanvas
import json
app = Dash(__name__)
filename = 'https://raw.githubusercontent.com/plotly/datasets/master/mitochondria.jpg'
canvas_width = 500
columns = ['type', 'width', 'height', 'scaleX', 'strokeWidth', 'path']
app.layout = html.Div([
html.H6('Draw on image and press Save to show annotations geometry'),
DashCanvas(id='annot-canvas',
lineWidth=5,
filename=filename,
width=canvas_width,
),
dash_table.DataTable(id='canvaas-table',
style_cell={'textAlign': 'left'},
columns=[{"name": i, "id": i} for i in columns]),
])
@callback(Output('canvaas-table', 'data'),
Input('annot-canvas', 'json_data'))
def update_data(string):
if string:
data = json.loads(string)
else:
raise PreventUpdate
return data['objects'][1:]
if __name__ == '__main__':
app.run(debug=True)
Draw on image and press Save to show annotations geometry
You can either write custom functions to parse the JSON string, or use the utility functions included in the dash_canvas package. In particular, dash_canvas.utils.parse_json_string returns a binary mask with non-zero pixels displaying the annotations:
import numpy as np
from skimage import io
from dash import Dash, html, Input, Output, callback
from dash.exceptions import PreventUpdate
from dash_canvas import DashCanvas
from dash_canvas.utils import array_to_data_url, parse_jsonstring
app = Dash(__name__)
filename = 'https://raw.githubusercontent.com/plotly/datasets/master/mitochondria.jpg'
canvas_width = 300
app.layout = html.Div([
html.H6('Draw on image and press Save to show annotations geometry'),
html.Div([
DashCanvas(
id='canvas',
lineWidth=5,
filename=filename,
width=canvas_width,
),
], className="five columns"),
html.Div(html.Img(id='my-iimage', width=300), className="five columns"),
])
@callback(Output('my-iimage', 'src'),
Input('canvas', 'json_data'))
def update_data(string):
if string:
mask = parse_jsonstring(string, io.imread(filename, as_gray=True).shape)
else:
raise PreventUpdate
return array_to_data_url((255 * mask).astype(np.uint8))
if __name__ == '__main__':
app.run(debug=True)
Draw on image and press Save to show annotations geometry
The above example uses the array_to_data_url utility function to transform a NumPy array into an image data string.
Finally, dash-canvas provides utility functions to process images given the binary mask derived from annotations:
from dash import Dash, html, Input, Output, callback
from dash.exceptions import PreventUpdate
from dash_canvas import DashCanvas
from dash_canvas.utils import (array_to_data_url, parse_jsonstring,
watershed_segmentation)
from skimage import io, color, img_as_ubyte
app = Dash(__name__)
filename = 'https://raw.githubusercontent.com/plotly/datasets/master/mitochondria.jpg'
canvas_width = 300
app.layout = html.Div([
html.H6('Annotate the two objects and the background'),
html.Div([
DashCanvas(
id='segmentation-canvas',
lineWidth=5,
filename=filename,
width=canvas_width,
),
], className="five columns"),
html.Div(html.Img(id='segmentation-iimg', width=300), className="five columns"),
])
@callback(Output('segmentation-iimg', 'src'),
Input('segmentation-canvas', 'json_data'))
def segmentation(string):
if string:
mask = parse_jsonstring(string, io.imread(filename, as_gray=True).shape)
seg = watershed_segmentation(io.imread(filename, as_gray=True), mask)
src = color.label2rgb(seg, image=io.imread(filename, as_gray=True))
else:
raise PreventUpdate
return array_to_data_url(img_as_ubyte(src))
if __name__ == '__main__':
app.run(debug=True)
Annotate the two objects and the background
These functions rely on scikit-image to process arrays as images. Here we used the watershed algorithm from scikit-image.
Updating the background image
The background image can be updated thanks to the image_content property (a str), for example using the contents property of dcc.Upload (an “open file” dialog). Updating image_content triggers the update of the json_data property containing the annotations.
More examples
A gallery of examples using DashCanvas is available at plotly/canvas-portal.