GUI system¶

Taichi has a built-in GUI system to help users visualize results.

Create a window¶

ti.GUI(title = 'Taichi', res = (512, 512), background_color = 0x000000, show_gui = True, fullscreen = False)¶

Parameters:	title – (optional, string) the window title res – (optional, scalar or tuple) resolution / size of the window background_color – (optional, RGB hex) background color of the window show_gui – (optional, bool) see the note below fullscreen – (optional, bool) `True` for fullscreen window fast_gui – (optional, bool) see Zero-copying frame buffer
Returns:	(GUI) an object represents the window

Create a window. If res is scalar, then width will be equal to height.

The following code creates a window of resolution 640x360:

gui = ti.GUI('Window Title', (640, 360))

Note

If you are running Taichi on a machine without a GUI environment, consider setting show_gui to False:

gui = ti.GUI('Window Title', (640, 360), show_gui=False)

while gui.running:
    ...
    gui.show(f'{gui.frame:06d}.png')  # save a series of screenshot

gui.show(filename = None)¶

Parameters:	gui – (GUI) the window object filename – (optional, string) see notes below

Show the window on the screen.

Note

If filename is specified, a screenshot will be saved to the file specified by the name. For example, the following saves frames of the window to .png’s:

for frame in range(10000):
    render(img)
    gui.set_image(img)
    gui.show(f'{frame:06d}.png')

Paint on a window¶

Taichi’s GUI supports painting simple geometric objects, such as lines, triangles, rectangles, circles, and text.

Note

The position parameter pos expects an input of a 2-element tuple, whose values are the relative position of the object. (0.0, 0.0) stands for the lower left corner of the window, and (1.0, 1.0) stands for the upper right corner.

gui.set_image(img)¶

Parameters:	gui – (GUI) the window object img – (np.array or ti.field) field containing the image, see notes below

Set an image to display on the window.

The image pixels are set from the values of img[i, j], where i indicates the horizontal coordinates (from left to right) and j the vertical coordinates (from bottom to top).

If the window size is (x, y), then img must be one of:

ti.field(shape=(x, y)), a grey-scale image
ti.field(shape=(x, y, 3)), where 3 is for (r, g, b) channels
ti.field(shape=(x, y, 2)), where 2 is for (r, g) channels
ti.Vector.field(3, shape=(x, y)) (r, g, b) channels on each component (see Vectors)
ti.Vector.field(2, shape=(x, y)) (r, g) channels on each component
np.ndarray(shape=(x, y))
np.ndarray(shape=(x, y, 3))
np.ndarray(shape=(x, y, 2))

The data type of img must be one of:

uint8, range [0, 255]
uint16, range [0, 65535]
uint32, range [0, 4294967295]
float32, range [0, 1]
float64, range [0, 1]

Note

When using float32 or float64 as the data type, img entries will be clipped into range [0, 1] for display.

gui.get_image()¶

Returns:	(np.array) the current image shown on the GUI

Get the 4-channel (RGBA) image shown in the current GUI system.

gui.circle(pos, color = 0xFFFFFF, radius = 1)¶

Parameters:	gui – (GUI) the window object pos – (tuple of 2) the position of the circle color – (optional, RGB hex) the color to fill the circle radius – (optional, scalar) the radius of the circle

Draw a solid circle.

gui.circles(pos, color = 0xFFFFFF, radius = 1)¶

Parameters:	gui – (GUI) the window object pos – (np.array) the positions of the circles color – (optional, RGB hex or np.array of uint32) the color(s) to fill the circles radius – (optional, scalar or np.array of float32) the radius (radii) of the circles

Draw solid circles.

Note

If color is a numpy array, the circle at pos[i] will be colored with color[i]. In this case, color must have the same size as pos.

gui.line(begin, end, color = 0xFFFFFF, radius = 1)¶

Parameters:	gui – (GUI) the window object begin – (tuple of 2) the first end point position of line end – (tuple of 2) the second end point position of line color – (optional, RGB hex) the color of line radius – (optional, scalar) the width of line

Draw a line.

gui.lines(begin, end, color = 0xFFFFFF, radius = 1)¶

Parameters:	gui – (GUI) the window object begin – (np.array) the positions of the first end point of lines end – (np.array) the positions of the second end point of lines color – (optional, RGB hex or np.array of uint32) the color(s) of lines radius – (optional, scalar or np.array of float32) the width(s) of the lines

Draw lines.

gui.triangle(a, b, c, color = 0xFFFFFF)¶

Parameters:	gui – (GUI) the window object a – (tuple of 2) the first end point position of triangle b – (tuple of 2) the second end point position of triangle c – (tuple of 2) the third end point position of triangle color – (optional, RGB hex) the color to fill the triangle

Draw a solid triangle.

gui.triangles(a, b, c, color = 0xFFFFFF)¶

Parameters:	gui – (GUI) the window object a – (np.array) the positions of the first end point of triangles b – (np.array) the positions of the second end point of triangles c – (np.array) the positions of the third end point of triangles color – (optional, RGB hex or np.array of uint32) the color(s) to fill the triangles

Draw solid triangles.

gui.rect(topleft, bottomright, radius = 1, color = 0xFFFFFF)¶

Parameters:	gui – (GUI) the window object topleft – (tuple of 2) the top-left point position of rectangle bottomright – (tuple of 2) the bottom-right point position of rectangle color – (optional, RGB hex) the color of stroke line radius – (optional, scalar) the width of stroke line

Draw a hollow rectangle.

gui.text(content, pos, font_size = 15, color = 0xFFFFFF)¶

Parameters:	gui – (GUI) the window object content – (str) the text to draw pos – (tuple of 2) the top-left point position of the fonts / texts font_size – (optional, scalar) the size of font (in height) color – (optional, RGB hex) the foreground color of text

Draw a line of text on screen.

ti.rgb_to_hex(rgb):

Parameters:	rgb – (tuple of 3 floats) The (R, G, B) float values, in range [0, 1]
Returns:	(RGB hex or np.array of uint32) The converted hex value

Convert a (R, G, B) tuple of floats into a single integer value. E.g.,

rgb = (0.4, 0.8, 1.0)
hex = ti.rgb_to_hex(rgb)  # 0x66ccff

rgb = np.array([[0.4, 0.8, 1.0], [0.0, 0.5, 1.0]])
hex = ti.rgb_to_hex(rgb)  # np.array([0x66ccff, 0x007fff])

The return values can be used in GUI drawing APIs.

Event processing¶

Every event have a key and type.

Event type is the type of event, for now, there are just three type of event:

ti.GUI.RELEASE  # key up or mouse button up
ti.GUI.PRESS    # key down or mouse button down
ti.GUI.MOTION   # mouse motion or mouse wheel

Event key is the key that you pressed on keyboard or mouse, can be one of:

# for ti.GUI.PRESS and ti.GUI.RELEASE event:
ti.GUI.ESCAPE  # Esc
ti.GUI.SHIFT   # Shift
ti.GUI.LEFT    # Left Arrow
'a'            # we use lowercase for alphabet
'b'
...
ti.GUI.LMB     # Left Mouse Button
ti.GUI.RMB     # Right Mouse Button

# for ti.GUI.MOTION event:
ti.GUI.MOVE    # Mouse Moved
ti.GUI.WHEEL   # Mouse Wheel Scrolling

A event filter is a list combined of key, type and (type, key) tuple, e.g.:

# if ESC pressed or released:
gui.get_event(ti.GUI.ESCAPE)

# if any key is pressed:
gui.get_event(ti.GUI.PRESS)

# if ESC pressed or SPACE released:
gui.get_event((ti.GUI.PRESS, ti.GUI.ESCAPE), (ti.GUI.RELEASE, ti.GUI.SPACE))

gui.running¶

Parameters:	gui – (GUI)
Returns:	(bool) `True` if `ti.GUI.EXIT` event occurred, vice versa

ti.GUI.EXIT occurs when you click on the close (X) button of a window. So gui.running will obtain False when the GUI is being closed.

For example, loop until the close button is clicked:

while gui.running:
    render()
    gui.set_image(pixels)
    gui.show()

You can also close the window by manually setting gui.running to False:

while gui.running:
    if gui.get_event(ti.GUI.ESCAPE):
        gui.running = False

    render()
    gui.set_image(pixels)
    gui.show()

gui.get_event(a, ...)¶

Parameters:	gui – (GUI) a – (optional, EventFilter) filter out matched events
Returns:	(bool) `False` if there is no pending event, vise versa

Try to pop a event from the queue, and store it in gui.event.

For example:

if gui.get_event():
    print('Got event, key =', gui.event.key)

For example, loop until ESC is pressed:

gui = ti.GUI('Title', (640, 480))
while not gui.get_event(ti.GUI.ESCAPE):
    gui.set_image(img)
    gui.show()

gui.get_events(a, ...)¶

Parameters:	gui – (GUI) a – (optional, EventFilter) filter out matched events
Returns:	(generator) a python generator, see below

Basically the same as gui.get_event, except for this one returns a generator of events instead of storing into gui.event:

for e in gui.get_events():
    if e.key == ti.GUI.ESCAPE:
        exit()
    elif e.key == ti.GUI.SPACE:
        do_something()
    elif e.key in ['a', ti.GUI.LEFT]:
        ...

gui.is_pressed(key, ...)¶

Parameters:	gui – (GUI) key – (EventKey) keys you want to detect
Returns:	(bool) `True` if one of the keys pressed, vice versa

Warning

Must be used together with gui.get_event, or it won’t be updated! For example:

while True:
    gui.get_event()  # must be called before is_pressed
    if gui.is_pressed('a', ti.GUI.LEFT):
        print('Go left!')
    elif gui.is_pressed('d', ti.GUI.RIGHT):
        print('Go right!')

gui.get_cursor_pos()¶

Parameters:	gui – (GUI)
Returns:	(tuple of 2) current cursor position within the window

For example:

mouse_x, mouse_y = gui.get_cursor_pos()

gui.fps_limit¶

Parameters:	gui – (GUI)
Returns:	(scalar or None) the maximum FPS, `None` for no limit

The default value is 60.

For example, to restrict FPS to be below 24, simply gui.fps_limit = 24. This helps reduce the overload on your hardware especially when you’re using OpenGL on your intergrated GPU which could make desktop slow to response.

GUI Widgets¶

Sometimes it’s more intuitive to use widgets like slider, button to control program variables instead of chaotic keyboard bindings. Taichi GUI provides a set of widgets that hopefully could make variable control more intuitive:

gui.slider(text, minimum, maximum, step=1)¶

Parameters:	text – (str) the text to be displayed above this slider. minumum – (float) the minimum value of the slider value. maxumum – (float) the maximum value of the slider value. step – (optional, float) the step between two separate value.
Returns:	(WidgetValue) a value getter / setter, see `WidgetValue`.

The widget will be display as: {text}: {value:.3f}, followed with a slider.

gui.label(text)¶

Parameters:	text – (str) the text to be displayed in the label.
Returns:	(WidgetValue) a value getter / setter, see `WidgetValue`.

The widget will be display as: {text}: {value:.3f}.

gui.button(text, event_name=None)¶

Parameters:	text – (str) the text to be displayed in the button. event_name – (optional, str) customize the event name.
Returns:	(EventKey) the event key for this button, see Event processing.

class WidgetValue¶

A getter / setter for widget values.

value¶: Get / set the current value in the widget where we’re returned from.

For example:

radius = gui.slider('Radius', 1, 50)

while gui.running:
    print('The radius now is', radius.value)
    ...
    radius.value += 0.01
    ...
    gui.show()

Image I/O¶

ti.imwrite(img, filename)¶

Parameters:	img – (ti.Vector.field or ti.field) the image you want to export filename – (string) the location you want to save to

Export a np.ndarray or Taichi field (ti.Matrix.field, ti.Vector.field, or ti.field) to a specified location filename.

Same as ti.GUI.show(filename), the format of the exported image is determined by the suffix of filename as well. Now ti.imwrite supports exporting images to png, img and jpg and we recommend using png.

Please make sure that the input image has a valid shape. If you want to export a grayscale image, the input shape of field should be (height, weight) or (height, weight, 1). For example:

import taichi as ti

ti.init()

shape = (512, 512)
type = ti.u8
pixels = ti.field(dtype=type, shape=shape)

@ti.kernel
def draw():
    for i, j in pixels:
        pixels[i, j] = ti.random() * 255    # integars between [0, 255] for ti.u8

draw()

ti.imwrite(pixels, f"export_u8.png")

Besides, for RGB or RGBA images, ti.imwrite needs to receive a field which has shape (height, width, 3) and (height, width, 4) individually.

Generally the value of the pixels on each channel of a png image is an integar in [0, 255]. For this reason, ti.imwrite will cast fields which has different datatypes all into integars between [0, 255]. As a result, ti.imwrite has the following requirements for different datatypes of input fields:

For float-type (ti.f16, ti.f32, etc) input fields, the value of each pixel should be float between [0.0, 1.0]. Otherwise ti.imwrite will first clip them into [0.0, 1.0]. Then they are multiplied by 256 and casted to integaters ranging from [0, 255].
For int-type (ti.u8, ti.u16, etc) input fields, the value of each pixel can be any valid integer in its own bounds. These integers in this field will be scaled to [0, 255] by being divided over the upper bound of its basic type accordingly.

Here is another example:

import taichi as ti

ti.init()

shape = (512, 512)
channels = 3
type = ti.f32
pixels = ti.Matrix.field(channels, dtype=type, shape=shape)

@ti.kernel
def draw():
    for i, j in pixels:
        for k in ti.static(range(channels)):
            pixels[i, j][k] = ti.random()   # floats between [0, 1] for ti.f32

draw()

ti.imwrite(pixels, f"export_f32.png")

ti.imread(filename, channels=0)¶

Parameters:	filename – (string) the filename of the image to load channels – (optional int) the number of channels in your specified image. The default value `0` means the channels of the returned image is adaptive to the image file
Returns:	(np.ndarray) the image read from `filename`

This function loads an image from the target filename and returns it as a np.ndarray(dtype=np.uint8).

Each value in this returned field is an integer in [0, 255].

ti.imshow(img, windname)¶

Parameters:	img – (ti.Vector.field or ti.field) the image to show in the GUI windname – (string) the name of the GUI window

This function will create an instance of ti.GUI and show the input image on the screen.

It has the same logic as ti.imwrite for different datatypes.

ti.imresize(img, w, h=None):

Parameters:	img – (np.array or ti.field) the input image. w – (int) the width after resizing. h – (optional, int) the height after resizing.
Returns:	(np.array) the resized image.

If h is not specified, it will be equal to w by default.

The output image shape is: (w, h, *img.shape[2:]).

Zero-copying frame buffer¶

When the GUI resolution (window size) is large, it sometimes becomes difficult to achieve 60 FPS even without any kernel invocations between two frames.

This is mainly due to the copy overhead, where Taichi GUI needs to copy the image buffer from one place to another. This copying is necessary for the 2D drawing functions, such as gui.circles, to work. The larger the image shape is, the larger the overhead.

Fortunately, sometimes your program only needs gui.set_image alone. In such cases, you can enable the fast_gui option for better performance. This mode allows Taichi GUI to directly write the image data to the frame buffer without additional copying, resulting in a much better FPS.

gui = ti.GUI(res, title, fast_gui=True)

Note

Because of the zero-copying mechanism, the image passed into gui.set_image must already be in the display-compatible format. That is, this field must either be a ti.Vector(3) (RGB) or a ti.Vector(4) (RGBA). In addition, each channel must be of type ti.f32, ti.f64 or ti.u8.

Note

If possible, consider enabling this option, especially when fullscreen=True.

Warning

Despite the performance boost, it has many limitations as trade off:

gui.set_image is the only available paint API in this mode. All other APIs like ``gui.circles``, ``gui.rect``, ``gui.triangles``, etc., won’t work.

gui.set_image will only takes Taichi 3D or 4D vector fields (RGB or RGBA) as input.