Source code for ginga.ImageView

#
# ImageView.py -- base class for the display of image files
#
# This is open-source software licensed under a BSD license.
# Please see the file LICENSE.txt for details.
#
"""This module handles image viewers."""

from io import BytesIO

import math
import logging
import threading
import sys
import traceback
import time
import uuid

import numpy as np

from ginga.misc import Callback, Settings
from ginga import BaseImage, AstroImage
from ginga import RGBMap, AutoCuts, ColorDist, zoom
from ginga import colors, trcalc
from ginga.canvas import coordmap, transform
from ginga.canvas.types.layer import DrawingCanvas
from ginga.util import addons, vip

__all__ = ['ImageViewBase']


class ImageViewError(Exception):
    pass


class ImageViewCoordsError(ImageViewError):
    pass


class ImageViewNoDataError(ImageViewError):
    pass


[docs]class ImageViewBase(Callback.Callbacks): """An abstract base class for displaying images represented by Numpy data arrays. This class attempts to do as much of the image handling as possible using Numpy array manipulations (even color and intensity mapping) so that only a minimal mapping to a pixel buffer is necessary in concrete subclasses that connect to an actual rendering surface (e.g., Qt, GTK, Tk, HTML5). Parameters ---------- logger : :py:class:`~logging.Logger` or `None` Logger for tracing and debugging. If not given, one will be created. rgbmap : `~ginga.RGBMap.RGBMapper` or `None` RGB mapper object. If not given, one will be created. settings : `~ginga.misc.Settings.SettingGroup` or `None` Viewer preferences. If not given, one will be created. """ vname = 'Ginga Image' vtypes = [BaseImage.BaseImage]
[docs] @classmethod def viewable(cls, dataobj): """Test whether `dataobj` is viewable by this viewer.""" if not isinstance(dataobj, BaseImage.BaseImage): return False shp = list(dataobj.shape) if len(shp) < 2: return False return True
def __init__(self, logger=None, rgbmap=None, settings=None): Callback.Callbacks.__init__(self) if logger is not None: self.logger = logger else: self.logger = logging.Logger('ImageViewBase') # Create settings and set defaults if settings is None: settings = Settings.SettingGroup(logger=self.logger) self.settings = settings # to be eventually deprecated self.t_ = settings # RGB mapper if rgbmap: # which way should the settings be migrated-- # rgbmap to viewer or vice-versa? t_ = rgbmap.get_settings() t_.share_settings(self.t_, keylist=rgbmap.settings_keys) else: rgbmap = RGBMap.RGBMapper(self.logger, settings=self.t_) self.rgbmap = rgbmap # Renderer self.renderer = None # for debugging self.viewer_id = str(uuid.uuid4()) self.name = self.viewer_id # Initialize RGBMap rgbmap.add_callback('changed', self.rgbmap_cb) # for scale self.t_.add_defaults(scale=(1.0, 1.0)) for name in ['scale']: self.t_.get_setting(name).add_callback('set', self.scale_cb) # for pan self.t_.add_defaults(pan=(1.0, 1.0), pan_coord='data') for name in ['pan', ]: # 'pan_coord' self.t_.get_setting(name).add_callback('set', self.pan_cb) # for cut levels self.t_.add_defaults(cuts=(0.0, 0.0)) for name in ['cuts']: self.t_.get_setting(name).add_callback('set', self.cut_levels_cb) # for auto cut levels self.autocuts_options = ('on', 'override', 'once', 'off') self.t_.add_defaults(autocuts='override', autocut_method='zscale', autocut_params=[]) for name in ('autocut_method', 'autocut_params'): self.t_.get_setting(name).add_callback('set', self.auto_levels_cb) # for zooming self.t_.add_defaults(zoomlevel=1.0, zoom_algorithm='step', scale_x_base=1.0, scale_y_base=1.0, interpolation='basic', zoom_rate=math.sqrt(2.0)) for name in ('zoom_rate', 'zoom_algorithm', 'scale_x_base', 'scale_y_base'): self.t_.get_setting(name).add_callback('set', self.zoomsetting_change_cb) self.zoom = zoom.get_zoom_alg(self.t_['zoom_algorithm'])(self) self.t_.get_setting('interpolation').add_callback( 'set', self.interpolation_change_cb) # max/min scaling self.t_.add_defaults(scale_max=None, scale_min=None) # autozoom options self.autozoom_options = ('on', 'override', 'once', 'off') self.t_.add_defaults(autozoom='on') # for panning self.autocenter_options = ('on', 'override', 'once', 'off') self.t_.add_defaults(autocenter='on') # for transforms self.t_.add_defaults(flip_x=False, flip_y=False, swap_xy=False) for name in ('flip_x', 'flip_y', 'swap_xy'): self.t_.get_setting(name).add_callback('set', self.transform_cb) # desired rotation angle self.t_.add_defaults(rot_deg=0.0) self.t_.get_setting('rot_deg').add_callback( 'set', self.rotation_change_cb) # misc self.t_.add_defaults(auto_orient=True, defer_redraw=True, defer_lagtime=0.025, show_pan_position=False, show_mode_indicator=True, show_focus_indicator=None, onscreen_font='Sans Serif', onscreen_font_size=None, color_fg="#D0F0E0", color_bg="#404040", limits=None, enter_focus=None) self.t_.get_setting('limits').add_callback('set', self._set_limits_cb) # embedded image "profiles" self.t_.add_defaults(profile_use_scale=False, profile_use_pan=False, profile_use_cuts=False, profile_use_transform=False, profile_use_rotation=False, profile_use_color_map=False) # ICC profile support d = dict(icc_output_profile=None, icc_output_intent='perceptual', icc_proof_profile=None, icc_proof_intent='perceptual', icc_black_point_compensation=False) self.t_.add_defaults(**d) for key in d: self.t_.get_setting(key).add_callback('set', self.icc_profile_cb) # viewer profile support self.use_image_profile = False self.profile_keylist = ['flip_x', 'flip_y', 'swap_xy', 'scale', 'pan', 'pan_coord', 'rot_deg', 'cuts', 'color_algorithm', 'color_hashsize', 'color_map', 'intensity_map', 'color_array', 'shift_array'] for name in self.t_.keys(): self.t_.get_setting(name).add_callback('set', self._update_profile_cb) # Object that calculates auto cut levels name = self.t_.get('autocut_method', 'zscale') klass = AutoCuts.get_autocuts(name) self.autocuts = klass(self.logger) self.vip = vip.ViewerImageProxy(self) # PRIVATE IMPLEMENTATION STATE # flag indicating whether our size has been set self._imgwin_set = False self._imgwin_wd = 0 self._imgwin_ht = 0 # desired size # on gtk, this seems to set a boundary on the lower size, so we # default to very small, set it larger with set_desired_size() #self._desired_size = (300, 300) self._desired_size = (1, 1) # viewer window backend has its canvas origin (0, 0) in upper left self.origin_upper = True # offset of pixel 0 from data coordinates # (pixels are centered on the coordinate) self.data_off = 0.5 self._invert_y = True # optimization of redrawing self.defer_redraw = self.t_.get('defer_redraw', True) self.defer_lagtime = self.t_.get('defer_lagtime', 0.025) self.time_last_redraw = time.time() self._defer_whence = 0 self._defer_whence_reset = 5 self._defer_lock = threading.RLock() self._defer_flag = False self._hold_redraw_cnt = 0 self.suppress_redraw = SuppressRedraw(self) # last known window mouse position self.last_win_x = 0 self.last_win_y = 0 # last known data mouse position self.last_data_x = 0 self.last_data_y = 0 self.orient_map = { # tag: (flip_x, flip_y, swap_xy) 1: (False, True, False), 2: (True, True, False), 3: (True, False, False), 4: (False, False, False), 5: (True, False, True), 6: (True, True, True), 7: (False, True, True), 8: (False, False, True), } # our canvas self.canvas = DrawingCanvas() self.canvas.initialize(None, self, self.logger) self.canvas.add_callback('modified', self.canvas_changed_cb) self.canvas.set_surface(self) self.canvas.ui_set_active(True, viewer=self) # private canvas for drawing self.private_canvas = self.canvas # handle to image object on the image canvas self._imgobj = None self._canvas_img_tag = '__image' # set up basic transforms self.trcat = transform.get_catalog() self.tform = {} self.recalc_transforms(self.trcat) self.coordmap = { 'native': coordmap.NativeMapper(self), 'window': coordmap.WindowMapper(self), 'percentage': coordmap.PercentageMapper(self), 'cartesian': coordmap.CartesianMapper(self), 'data': coordmap.DataMapper(self), None: coordmap.DataMapper(self), 'offset': coordmap.OffsetMapper(self, None), 'wcs': coordmap.WCSMapper(self), } # cursors self.cursor = {} # setup default fg color color = self.t_.get('color_fg', "#D0F0E0") r, g, b = colors.lookup_color(color) self.img_fg = (r, g, b) # setup default bg color color = self.t_.get('color_bg', "#404040") r, g, b = colors.lookup_color(color) self.img_bg = (r, g, b) # For callbacks for name in ('transform', 'image-set', 'image-unset', 'configure', 'redraw', 'limits-set', 'cursor-changed'): self.enable_callback(name) # for timed refresh self.rf_fps = 1 self.rf_rate = 1.0 / self.rf_fps self.rf_timer = self.make_timer() self.rf_flags = {} self.rf_draw_count = 0 self.rf_delta_total = 0.0 self.rf_timer_count = 0 self.rf_start_time = 0.0 self.rf_late_warn_time = 0.0 self.rf_late_warn_interval = 10.0 self.rf_late_total = 0.0 self.rf_late_count = 0 self.rf_early_total = 0.0 self.rf_early_count = 0 self.rf_skip_total = 0.0 if self.rf_timer is not None: self.rf_timer.add_callback('expired', self.refresh_timer_cb, self.rf_flags)
[docs] def set_window_size(self, width, height): """Report the size of the window to display the image. **Callbacks** Will call any callbacks registered for the ``'configure'`` event. Callbacks should have a method signature of:: (viewer, width, height, ...) .. note:: This is called by the subclass with ``width`` and ``height`` as soon as the actual dimensions of the allocated window are known. Parameters ---------- width : int The width of the window in pixels. height : int The height of the window in pixels. """ width, height = int(width), int(height) self._imgwin_wd = width self._imgwin_ht = height self.logger.debug("widget resized to %dx%d" % (width, height)) self.renderer.resize((width, height)) self.make_callback('configure', width, height)
[docs] def configure(self, width, height): """See :meth:`set_window_size`.""" self._imgwin_set = True self.set_window_size(width, height)
[docs] def configure_surface(self, width, height): """See :meth:`configure`.""" # legacy API--to be deprecated self.configure(width, height)
[docs] def set_desired_size(self, width, height): """See :meth:`set_window_size`.""" self._desired_size = (width, height) if not self._imgwin_set: self.set_window_size(width, height)
[docs] def get_desired_size(self): """Get desired size. Returns ------- size : tuple Desired size in the form of ``(width, height)``. """ return self._desired_size
[docs] def get_window_size(self): """Get the window size in the underlying implementation. Returns ------- size : tuple Window size in the form of ``(width, height)``. """ return (self._imgwin_wd, self._imgwin_ht)
[docs] def get_dims(self, data): """Get the first two dimensions of Numpy array data. Data may have more dimensions, but they are not reported. Returns ------- dims : tuple Data dimensions in the form of ``(width, height)``. """ height, width = data.shape[:2] return (width, height)
[docs] def get_settings(self): """Get the settings used by this instance. Returns ------- settings : `~ginga.misc.Settings.SettingGroup` Settings. """ return self.t_
[docs] def get_logger(self): """Get the logger used by this instance. Returns ------- logger : :py:class:`~logging.Logger` Logger. """ return self.logger
[docs] def get_vip(self): """Get the ViewerImageProxy object used by this instance. Returns ------- vip : `~ginga.util.vip.ViewerImageProxy` A ViewerImageProxy object. """ return self.vip
[docs] def set_renderer(self, renderer): """Set and initialize the renderer used by this instance. """ self.renderer = renderer width, height = self.get_window_size() if width > 0 and height > 0: renderer.resize((width, height))
[docs] def get_canvas(self): """Get the canvas object used by this instance. Returns ------- canvas : `~ginga.canvas.types.layer.DrawingCanvas` Canvas. """ return self.canvas
[docs] def set_canvas(self, canvas, private_canvas=None): """Set the canvas object. Parameters ---------- canvas : `~ginga.canvas.types.layer.DrawingCanvas` Canvas object. private_canvas : `~ginga.canvas.types.layer.DrawingCanvas` or `None` Private canvas object. If not given, this is the same as ``canvas``. """ self.canvas = canvas canvas.initialize(None, self, self.logger) canvas.add_callback('modified', self.canvas_changed_cb) canvas.set_surface(self) canvas.ui_set_active(True, viewer=self) self._imgobj = None # private canvas set? if private_canvas is not None: self.private_canvas = private_canvas self.initialize_private_canvas(self.private_canvas) if private_canvas != canvas: private_canvas.set_surface(self) private_canvas.ui_set_active(True, viewer=self) private_canvas.add_callback('modified', self.canvas_changed_cb) # sanity check that we have a private canvas, and if not, # set it to the "advertised" canvas if self.private_canvas is None: self.private_canvas = canvas self.initialize_private_canvas(self.private_canvas) # make sure private canvas has our non-private one added if (self.private_canvas != self.canvas) and ( not self.private_canvas.has_object(canvas)): self.private_canvas.add(canvas)
[docs] def get_private_canvas(self): """Get the private canvas object used by this instance. Returns ------- canvas : `~ginga.canvas.types.layer.DrawingCanvas` Canvas. """ return self.private_canvas
[docs] def initialize_private_canvas(self, private_canvas): """Initialize the private canvas used by this instance. """ if self.t_.get('show_pan_position', False): self.show_pan_mark(True) if self.t_.get('show_focus_indicator', False): self.show_focus_indicator(True)
[docs] def set_color_map(self, cmap_name): """Set the color map. Available color map names can be discovered using :func:`~ginga.cmap.get_names`. Parameters ---------- cmap_name : str The name of a color map. """ self.t_.set(color_map=cmap_name)
[docs] def set_intensity_map(self, imap_name): """Set the intensity map. Available intensity map names can be discovered using :func:`ginga.imap.get_names`. Parameters ---------- imap_name : str The name of an intensity map. """ self.t_.set(intensity_map=imap_name)
[docs] def set_color_algorithm(self, calg_name, **kwdargs): """Set the color distribution algorithm. Available color distribution algorithm names can be discovered using :func:`ginga.ColorDist.get_dist_names`. Parameters ---------- calg_name : str The name of a color distribution algorithm. kwdargs : dict Keyword arguments for color distribution object (see `~ginga.ColorDist`). """ # TEMP: ignore kwdargs self.t_.set(color_algorithm=calg_name)
[docs] def get_color_algorithms(self): """Get available color distribution algorithm names. See :func:`ginga.ColorDist.get_dist_names`. """ return ColorDist.get_dist_names()
[docs] def set_cmap(self, cm): """Set color map. See :meth:`ginga.RGBMap.RGBMapper.set_cmap`. """ rgbmap = self.get_rgbmap() rgbmap.set_cmap(cm)
[docs] def invert_cmap(self): """Invert the color map. See :meth:`ginga.RGBMap.RGBMapper.invert_cmap`. """ rgbmap = self.get_rgbmap() rgbmap.invert_cmap()
[docs] def set_imap(self, im): """Set intensity map. See :meth:`ginga.RGBMap.RGBMapper.set_imap`. """ rgbmap = self.get_rgbmap() rgbmap.set_imap(im)
[docs] def set_calg(self, dist): """Set color distribution algorithm. See :meth:`ginga.RGBMap.RGBMapper.set_dist`. """ rgbmap = self.get_rgbmap() rgbmap.set_dist(dist)
[docs] def shift_cmap(self, pct): """Shift color map. See :meth:`ginga.RGBMap.RGBMapper.shift`. """ rgbmap = self.get_rgbmap() rgbmap.shift(pct)
[docs] def scale_and_shift_cmap(self, scale_pct, shift_pct): """Stretch and/or shrink the color map. See :meth:`ginga.RGBMap.RGBMapper.scale_and_shift`. """ rgbmap = self.get_rgbmap() rgbmap.scale_and_shift(scale_pct, shift_pct)
[docs] def restore_contrast(self): """Restores the color map from any stretch and/or shrinkage. See :meth:`ginga.RGBMap.RGBMapper.reset_sarr`. """ rgbmap = self.get_rgbmap() rgbmap.reset_sarr()
[docs] def restore_cmap(self): """Restores the color map from any rotation, stretch and/or shrinkage. See :meth:`ginga.RGBMap.RGBMapper.restore_cmap`. """ rgbmap = self.get_rgbmap() rgbmap.restore_cmap()
[docs] def rgbmap_cb(self, rgbmap): """Handle callback for when RGB map has changed.""" self.logger.debug("RGB map has changed.") self.renderer.rgbmap_change(rgbmap)
[docs] def get_rgbmap(self): """Get the RGB map object used by this instance. Returns ------- rgbmap : `~ginga.RGBMap.RGBMapper` RGB map. """ return self.rgbmap
[docs] def set_rgbmap(self, rgbmap): """Set RGB map object used by this instance. It controls how the values in the image are mapped to color. Parameters ---------- rgbmap : `~ginga.RGBMap.RGBMapper` RGB map. """ self.rgbmap = rgbmap t_ = rgbmap.get_settings() t_.share_settings(self.t_, keylist=rgbmap.settings_keys) rgbmap.add_callback('changed', self.rgbmap_cb) self.renderer.rgbmap_change(rgbmap)
[docs] def get_image(self): """Get the image currently being displayed. Returns ------- image : `~ginga.AstroImage.AstroImage` or `~ginga.RGBImage.RGBImage` Image object. """ if self._imgobj is not None: # quick optimization return self._imgobj.get_image() canvas_img = self.get_canvas_image() return canvas_img.get_image()
# for compatibility with other viewers get_dataobj = get_image
[docs] def get_canvas_image(self): """Get canvas image object. Returns ------- imgobj : `~ginga.canvas.types.image.NormImage` Normalized image sitting on the canvas. """ if self._imgobj is not None: return self._imgobj try: # See if there is an image on the canvas self._imgobj = self.canvas.get_object_by_tag(self._canvas_img_tag) self._imgobj.add_callback('image-set', self._image_set_cb) except KeyError: # add a normalized image item to this canvas if we don't # have one already--then just keep reusing it NormImage = self.canvas.get_draw_class('normimage') self._imgobj = NormImage(0, 0, None, alpha=1.0, interpolation=None) self._imgobj.is_data = True self._imgobj.add_callback('image-set', self._image_set_cb) return self._imgobj
[docs] def set_image(self, image, add_to_canvas=True): """Set an image to be displayed. If there is no error, the ``'image-unset'`` and ``'image-set'`` callbacks will be invoked. Parameters ---------- image : `~ginga.AstroImage.AstroImage` or `~ginga.RGBImage.RGBImage` 2D Image object. add_to_canvas : bool Add image to canvas. """ if not self.viewable(image): raise ValueError("Wrong type of object to load: %s" % ( str(type(image)))) canvas_img = self.get_canvas_image() old_image = canvas_img.get_image() self.make_callback('image-unset', old_image) with self.suppress_redraw: # update viewer limits wd, ht = image.get_size() limits = ((-self.data_off, -self.data_off), (float(wd - self.data_off), float(ht - self.data_off))) self.t_.set(limits=limits) # this line should force the callback of _image_set_cb() canvas_img.set_image(image) if add_to_canvas: try: self.canvas.get_object_by_tag(self._canvas_img_tag) except KeyError: self.canvas.add(canvas_img, tag=self._canvas_img_tag) # move image to bottom of layers self.canvas.lower_object(canvas_img) self.canvas.update_canvas(whence=0)
# for compatibility with other viewers set_dataobj = set_image def _image_set_cb(self, canvas_img, image): try: self.apply_profile_or_settings(image) except Exception as e: self.logger.error("Failed to initialize image: %s" % (str(e))) try: # log traceback, if possible (type, value, tb) = sys.exc_info() tb_str = "".join(traceback.format_tb(tb)) self.logger.error("Traceback:\n%s" % (tb_str)) except Exception: tb_str = "Traceback information unavailable." self.logger.error(tb_str) # update our display if the image changes underneath us image.add_callback('modified', self._image_modified_cb) # out with the old, in with the new... self.make_callback('image-set', image)
[docs] def apply_profile_or_settings(self, image): """Apply a profile to the viewer. Parameters ---------- image : `~ginga.AstroImage.AstroImage` or `~ginga.RGBImage.RGBImage` Image object. This function is used to initialize the viewer when a new image is loaded. Either the embedded profile settings or the default settings are applied as specified in the channel preferences. """ profile = image.get('profile', None) keylist = [] with self.suppress_redraw: if profile is not None: if self.t_['profile_use_transform'] and 'flip_x' in profile: keylist.extend(['flip_x', 'flip_y', 'swap_xy']) if self.t_['profile_use_scale'] and 'scale' in profile: keylist.extend(['scale']) if self.t_['profile_use_pan'] and 'pan' in profile: keylist.extend(['pan']) if self.t_['profile_use_rotation'] and 'rot_deg' in profile: keylist.extend(['rot_deg']) if self.t_['profile_use_cuts'] and 'cuts' in profile: keylist.extend(['cuts']) if self.t_['profile_use_color_map'] and 'color_map' in profile: keylist.extend(['color_algorithm', 'color_hashsize', 'color_map', 'intensity_map', 'color_array', 'shift_array']) self.apply_profile(profile, keylist=keylist) # proceed with initialization that is not in the profile # initialize transforms if self.t_['auto_orient'] and 'flip_x' not in keylist: self.logger.debug( "auto orient (%s)" % (self.t_['auto_orient'])) self.auto_orient() # initialize scale if self.t_['autozoom'] != 'off' and 'scale' not in keylist: self.logger.debug("auto zoom (%s)" % (self.t_['autozoom'])) self.zoom_fit(no_reset=True) # initialize pan position # NOTE: False a possible value from historical use if (self.t_['autocenter'] not in ('off', False) and 'pan' not in keylist): self.logger.debug( "auto center (%s)" % (self.t_['autocenter'])) self.center_image(no_reset=True) # initialize cuts if self.t_['autocuts'] != 'off' and 'cuts' not in keylist: self.logger.debug("auto cuts (%s)" % (self.t_['autocuts'])) self.auto_levels() # save the profile in the image if self.use_image_profile: self.checkpoint_profile()
[docs] def apply_profile(self, profile, keylist=None): """Apply a profile to the viewer. Parameters ---------- profile : `~ginga.misc.Settings.SettingGroup` This function is used to initialize the viewer to a known state. The keylist, if given, will limit the items to be transferred from the profile to viewer settings, otherwise all items are copied. """ with self.suppress_redraw: profile.copy_settings(self.t_, keylist=keylist, callback=True)
[docs] def capture_profile(self, profile): self.t_.copy_settings(profile) self.logger.debug("profile attributes set")
[docs] def checkpoint_profile(self): profile = self.save_profile() self.capture_profile(profile) return profile
[docs] def save_profile(self, **params): """Save the given parameters into profile settings. Parameters ---------- params : dict Keywords and values to be saved. """ image = self.get_image() if (image is None): return profile = image.get('profile', None) if profile is None: # If image has no profile then create one profile = Settings.SettingGroup() image.set(profile=profile) self.logger.debug("saving to image profile: params=%s" % ( str(params))) profile.set(**params) return profile
def _update_profile_cb(self, setting, value): key = setting.name if self.use_image_profile: # ? and key in self.profile_keylist kwargs = {key: value} self.save_profile(**kwargs) def _image_modified_cb(self, image): canvas_img = self.get_canvas_image() image2 = canvas_img.get_image() if image is not image2: # not the image we are now displaying, perhaps a former image return with self.suppress_redraw: # update viewer limits wd, ht = image.get_size() limits = ((-self.data_off, -self.data_off), (float(wd - self.data_off), float(ht - self.data_off))) self.t_.set(limits=limits) canvas_img.reset_optimize() # Per issue #111, zoom and pan and cuts probably should # not change if the image is _modified_, or it should be # optional--these settings are only for _new_ images # UPDATE: don't zoom/pan (assuming image size, etc. hasn't # changed), but *do* apply cuts try: self.logger.debug("image data updated") ## if self.t_['auto_orient']: ## self.auto_orient() ## if self.t_['autozoom'] != 'off': ## self.zoom_fit(no_reset=True) ## if not self.t_['autocenter'] in ('off', False): ## self.center_image() if self.t_['autocuts'] != 'off': self.auto_levels() except Exception as e: self.logger.error("Failed to initialize image: %s" % (str(e))) try: # log traceback, if possible (type, value, tb) = sys.exc_info() tb_str = "".join(traceback.format_tb(tb)) self.logger.error("Traceback:\n%s" % (tb_str)) except Exception: tb_str = "Traceback information unavailable." self.logger.error(tb_str) self.canvas.update_canvas(whence=1)
[docs] def set_data(self, data, metadata=None): """Set an image to be displayed by providing raw data. This is a convenience method for first constructing an image with `~ginga.AstroImage.AstroImage` and then calling :meth:`set_image`. Parameters ---------- data : ndarray This should be at least a 2D Numpy array. metadata : dict or `None` Image metadata mapping keywords to their respective values. """ image = AstroImage.AstroImage(data, metadata=metadata, logger=self.logger) self.set_image(image)
[docs] def clear(self): """Clear the displayed image.""" self._imgobj = None try: # See if there is an image on the canvas self.canvas.delete_object_by_tag(self._canvas_img_tag) self.redraw() except KeyError: pass
[docs] def copy_to_dst(self, target): """Extract our image and call :meth:`set_image` on the target with it. Parameters ---------- target Subclass of `ImageViewBase`. """ image = self.get_image() target.set_image(image)
[docs] def redraw(self, whence=0): """Redraw the canvas. Parameters ---------- whence : int or float Optimization flag that reduces the time to refresh the viewer by only recalculating what is necessary: 0: New image, pan/scale has changed 1: Cut levels or similar has changed 2: Color mapping has changed 2.3: ICC profile has changed 2.5: Transforms have changed 2.6: Rotation has changed 3: Graphical overlays have changed """ with self._defer_lock: whence = min(self._defer_whence, whence) if not self.defer_redraw: if self._hold_redraw_cnt == 0: self._defer_whence = self._defer_whence_reset self.redraw_now(whence=whence) else: self._defer_whence = whence return elapsed = time.time() - self.time_last_redraw # If there is no redraw scheduled, or we are overdue for one: if (not self._defer_flag) or (elapsed > self.defer_lagtime): # If more time than defer_lagtime has passed since the # last redraw then just do the redraw immediately if elapsed > self.defer_lagtime: if self._hold_redraw_cnt > 0: #self._defer_flag = True self._defer_whence = whence return self._defer_whence = self._defer_whence_reset self.logger.debug("lagtime expired--forced redraw") self.redraw_now(whence=whence) return # Indicate that a redraw is necessary and record whence self._defer_flag = True self._defer_whence = whence # schedule a redraw by the end of the defer_lagtime secs = self.defer_lagtime - elapsed self.logger.debug("defer redraw (whence=%.2f) in %.f sec" % ( whence, secs)) self.reschedule_redraw(secs) else: # A redraw is already scheduled. Just record whence. self._defer_whence = whence self.logger.debug("update whence=%.2f" % (whence))
[docs] def is_redraw_pending(self): """Indicates whether a deferred redraw has been scheduled. Returns ------- pending : bool True if a deferred redraw is pending, False otherwise. """ return self._defer_whence < self._defer_whence_reset
[docs] def canvas_changed_cb(self, canvas, whence): """Handle callback for when canvas has changed.""" self.logger.debug("root canvas changed, whence=%d" % (whence)) # special check for whether image changed out from under us in # a shared canvas scenario try: # See if there is an image on the canvas canvas_img = self.canvas.get_object_by_tag(self._canvas_img_tag) if self._imgobj is not canvas_img: self._imgobj = canvas_img self._imgobj.add_callback('image-set', self._image_set_cb) self._image_set_cb(canvas_img, canvas_img.get_image()) except KeyError: self._imgobj = None self.redraw(whence=whence)
[docs] def delayed_redraw(self): """Handle delayed redrawing of the canvas.""" # This is the optimized redraw method with self._defer_lock: # pick up the lowest necessary level of redrawing whence = self._defer_whence self._defer_whence = self._defer_whence_reset flag = self._defer_flag self._defer_flag = False if flag: # If a redraw was scheduled, do it now self.redraw_now(whence=whence)
[docs] def set_redraw_lag(self, lag_sec): """Set lag time for redrawing the canvas. Parameters ---------- lag_sec : float Number of seconds to wait. """ self.defer_redraw = (lag_sec > 0.0) if self.defer_redraw: self.defer_lagtime = lag_sec
[docs] def set_refresh_rate(self, fps): """Set the refresh rate for redrawing the canvas at a timed interval. Parameters ---------- fps : float Desired rate in frames per second. """ self.rf_fps = fps self.rf_rate = 1.0 / self.rf_fps #self.set_redraw_lag(self.rf_rate) self.logger.info("set a refresh rate of %.2f fps" % (self.rf_fps))
[docs] def start_refresh(self): """Start redrawing the canvas at the previously set timed interval. """ self.logger.debug("starting timed refresh interval") self.rf_flags['done'] = False self.rf_draw_count = 0 self.rf_timer_count = 0 self.rf_late_count = 0 self.rf_late_total = 0.0 self.rf_early_count = 0 self.rf_early_total = 0.0 self.rf_delta_total = 0.0 self.rf_skip_total = 0.0 self.rf_start_time = time.time() self.rf_deadline = self.rf_start_time self.refresh_timer_cb(self.rf_timer, self.rf_flags)
[docs] def stop_refresh(self): """Stop redrawing the canvas at the previously set timed interval. """ self.logger.debug("stopping timed refresh") self.rf_flags['done'] = True self.rf_timer.clear()
[docs] def get_refresh_stats(self): """Return the measured statistics for timed refresh intervals. Returns ------- stats : float The measured rate of actual back end updates in frames per second. """ if self.rf_draw_count == 0: fps = 0.0 else: interval = time.time() - self.rf_start_time fps = self.rf_draw_count / interval jitter = self.rf_delta_total / max(1, self.rf_timer_count) late_avg = self.rf_late_total / max(1, self.rf_late_count) late_pct = self.rf_late_count / max(1.0, float(self.rf_timer_count)) * 100 early_avg = self.rf_early_total / max(1, self.rf_early_count) early_pct = self.rf_early_count / max(1.0, float(self.rf_timer_count)) * 100 balance = self.rf_late_total - self.rf_early_total stats = dict(fps=fps, jitter=jitter, early_avg=early_avg, early_pct=early_pct, late_avg=late_avg, late_pct=late_pct, balance=balance) return stats
[docs] def refresh_timer_cb(self, timer, flags): """Refresh timer callback. This callback will normally only be called internally. Parameters ---------- timer : a Ginga GUI timer A GUI-based Ginga timer flags : dict-like A set of flags controlling the timer """ # this is the timer call back, from the GUI thread start_time = time.time() if flags.get('done', False): return # calculate next deadline deadline = self.rf_deadline self.rf_deadline += self.rf_rate self.rf_timer_count += 1 delta = abs(start_time - deadline) self.rf_delta_total += delta adjust = 0.0 if start_time > deadline: # we are late self.rf_late_total += delta self.rf_late_count += 1 late_avg = self.rf_late_total / self.rf_late_count adjust = - (late_avg / 2.0) self.rf_skip_total += delta if self.rf_skip_total < self.rf_rate: self.rf_draw_count += 1 # TODO: can we optimize whence? self.redraw_now(whence=0) else: # <-- we are behind by amount of time equal to one frame. # skip a redraw and attempt to catch up some time self.rf_skip_total = 0 else: if start_time < deadline: # we are early self.rf_early_total += delta self.rf_early_count += 1 self.rf_skip_total = max(0.0, self.rf_skip_total - delta) early_avg = self.rf_early_total / self.rf_early_count adjust = early_avg / 4.0 self.rf_draw_count += 1 # TODO: can we optimize whence? self.redraw_now(whence=0) delay = max(0.0, self.rf_deadline - time.time() + adjust) timer.start(delay)
[docs] def redraw_now(self, whence=0): """Redraw the displayed image. Parameters ---------- whence See :meth:`redraw`. """ try: time_start = time.time() self.renderer.initialize() self.redraw_data(whence=whence) self.renderer.finalize() # finally update the window drawable from the offscreen surface self.update_widget() time_done = time.time() time_delta = time_start - self.time_last_redraw time_elapsed = time_done - time_start self.time_last_redraw = time_done self.logger.debug( "widget '%s' redraw (whence=%d) delta=%.4f elapsed=%.4f sec" % ( self.name, whence, time_delta, time_elapsed)) except Exception as e: self.logger.error("Error redrawing image: %s" % (str(e))) try: # log traceback, if possible (type, value, tb) = sys.exc_info() tb_str = "".join(traceback.format_tb(tb)) self.logger.error("Traceback:\n%s" % (tb_str)) except Exception: tb_str = "Traceback information unavailable." self.logger.error(tb_str)
[docs] def redraw_data(self, whence=0): """Render image from RGB map and redraw private canvas. .. note:: Do not call this method unless you are implementing a subclass. Parameters ---------- whence See :meth:`redraw`. """ if not self._imgwin_set: # window has not been realized yet return self._whence = whence self.renderer.render_whence(whence) self.private_canvas.draw(self) self.make_callback('redraw', whence) if whence < 3: self.check_cursor_location()
[docs] def check_cursor_location(self): """Check whether the data location of the last known position of the cursor has changed. If so, issue a callback. """ # Check whether cursor data position has changed relative # to previous value data_x, data_y = self.get_data_xy(self.last_win_x, self.last_win_y) if (data_x != self.last_data_x or data_y != self.last_data_y): self.last_data_x, self.last_data_y = data_x, data_y self.logger.debug("cursor location changed %.4f,%.4f => %.4f,%.4f" % ( self.last_data_x, self.last_data_y, data_x, data_y)) # we make this call compatible with the motion callback # for now, but there is no concept of a button here button = 0 self.make_ui_callback('cursor-changed', button, data_x, data_y) return data_x, data_y
[docs] def getwin_array(self, order='RGB', alpha=1.0, dtype=None): return self.renderer.getwin_array(order=order, alpha=alpha, dtype=dtype)
[docs] def getwin_buffer(self, order='RGB', alpha=1.0, dtype=None): """Same as :meth:`getwin_array`, but with the output array converted to C-order Python bytes. """ outarr = self.renderer.getwin_array(order=order, alpha=alpha, dtype=dtype) if not hasattr(outarr, 'tobytes'): # older versions of numpy return outarr.tostring(order='C') return outarr.tobytes(order='C')
[docs] def get_limits(self, coord='data'): """Get the bounding box of the viewer extents. Returns ------- limits : tuple Bounding box in coordinates of type `coord` in the form of ``(ll_pt, ur_pt)``. """ limits = self.t_['limits'] if limits is None: # No user defined limits. # Calculate limits based on plotted points, if any canvas = self.get_canvas() pts = canvas.get_points() if len(pts) > 0: limits = trcalc.get_bounds(pts) else: # No limits found, go to default limits = ((0.0, 0.0), (0.0, 0.0)) # convert to desired coordinates crdmap = self.get_coordmap(coord) limits = crdmap.data_to(limits) return limits
[docs] def set_limits(self, limits, coord='data'): """Set the bounding box of the viewer extents. Parameters ---------- limits : tuple or None A tuple setting the extents of the viewer in the form of ``(ll_pt, ur_pt)``. """ if limits is not None: if len(limits) != 2: raise ValueError("limits takes a 2 tuple, or None") # convert to data coordinates crdmap = self.get_coordmap(coord) limits = crdmap.to_data(limits) self.t_.set(limits=limits)
[docs] def reset_limits(self): """Reset the bounding box of the viewer extents. Parameters ---------- None """ self.t_.set(limits=None)
def _set_limits_cb(self, setting, limits): self.renderer.limits_change(limits) # TODO: deprecate this chained callback and have users just use # 'set' callback for "limits" setting ? self.make_callback('limits-set', limits)
[docs] def icc_profile_cb(self, setting, value): """Handle callback related to changes in output ICC profiles.""" self.renderer.icc_profile_change()
[docs] def get_data_pt(self, win_pt): """Similar to :meth:`get_data_xy`, except that it takes a single array of points. """ return self.tform['mouse_to_data'].to_(win_pt)
[docs] def get_data_xy(self, win_x, win_y, center=None): """Get the closest coordinates in the data array to those reported on the window. Parameters ---------- win_x, win_y : float or ndarray Window coordinates. center : bool If `True`, then the coordinates are mapped such that the pixel is centered on the square when the image is zoomed in past 1X. This is the specification of the FITS image standard, that the pixel is centered on the integer row/column. Returns ------- coord : tuple Data coordinates in the form of ``(x, y)``. """ if center is not None: self.logger.warning("`center` keyword is ignored and will be deprecated") arr_pts = np.asarray((win_x, win_y)).T return self.tform['mouse_to_data'].to_(arr_pts).T[:2]
[docs] def offset_to_data(self, off_x, off_y, center=None): """Get the closest coordinates in the data array to those in cartesian fixed (non-scaled) canvas coordinates. Parameters ---------- off_x, off_y : float or ndarray Cartesian canvas coordinates. Returns ------- coord : tuple Data coordinates in the form of ``(x, y)``. """ if center is not None: self.logger.warning("`center` keyword is ignored and will be deprecated") arr_pts = np.asarray((off_x, off_y)).T return self.tform['data_to_cartesian'].from_(arr_pts).T[:2]
[docs] def get_canvas_pt(self, data_pt): """Similar to :meth:`get_canvas_xy`, except that it takes a single array of points. """ return self.tform['data_to_native'].to_(data_pt)
[docs] def get_canvas_xy(self, data_x, data_y, center=None): """Reverse of :meth:`get_data_xy`. """ if center is not None: self.logger.warning("`center` keyword is ignored and will be deprecated") arr_pts = np.asarray((data_x, data_y)).T return self.tform['data_to_native'].to_(arr_pts).T[:2]
[docs] def data_to_offset(self, data_x, data_y, center=None): """Reverse of :meth:`offset_to_data`. """ if center is not None: self.logger.warning("`center` keyword is ignored and will be deprecated") arr_pts = np.asarray((data_x, data_y)).T return self.tform['data_to_cartesian'].to_(arr_pts).T[:2]
[docs] def offset_to_window(self, off_x, off_y): """Convert data offset to window coordinates. Parameters ---------- off_x, off_y : float or ndarray Data offsets. Returns ------- coord : tuple Offset in window coordinates in the form of ``(x, y)``. """ arr_pts = np.asarray((off_x, off_y)).T return self.tform['cartesian_to_native'].to_(arr_pts).T[:2]
[docs] def window_to_offset(self, win_x, win_y): """Reverse of :meth:`offset_to_window`.""" arr_pts = np.asarray((win_x, win_y)).T return self.tform['cartesian_to_native'].from_(arr_pts).T[:2]
[docs] def canvascoords(self, data_x, data_y, center=None): """Same as :meth:`get_canvas_xy`. """ if center is not None: self.logger.warning("`center` keyword is ignored and will be deprecated") # data->canvas space coordinate conversion arr_pts = np.asarray((data_x, data_y)).T return self.tform['data_to_native'].to_(arr_pts).T[:2]
[docs] def get_data_size(self): """Get the dimensions of the image currently being displayed. Returns ------- size : tuple Image dimensions in the form of ``(width, height)``. """ xy_mn, xy_mx = self.get_limits() ht = abs(xy_mx[1] - xy_mn[1]) wd = abs(xy_mx[0] - xy_mn[0]) return (wd, ht)
[docs] def get_data_pct(self, xpct, ypct): """Calculate new data size for the given axis ratios. See :meth:`get_limits`. Parameters ---------- xpct, ypct : float Ratio for X and Y, respectively, where 1 is 100%. Returns ------- x, y : int Scaled dimensions. """ wd, ht = self.get_data_size() x, y = int(float(xpct) * wd), int(float(ypct) * ht) return (x, y)
[docs] def get_pan_rect(self): """Get the coordinates in the actual data corresponding to the area shown in the display for the current zoom level and pan. Returns ------- points : list Coordinates in the form of ``[(x0, y0), (x1, y1), (x2, y2), (x3, y3)]`` from lower-left to lower-right. """ wd, ht = self.get_window_size() win_pts = np.asarray([(0, 0), (wd, 0), (wd, ht), (0, ht)]) arr_pts = self.tform['data_to_window'].from_(win_pts) return arr_pts
[docs] def get_draw_rect(self): """Get the coordinates in the actual data corresponding to the area needed for drawing images for the current zoom level and pan. Unlike get_pan_rect(), this includes areas outside of the current viewport, but that might be viewed with a transformation or rotation subsequently applied. Returns ------- points : list Coordinates in the form of ``[(x0, y0), (x1, y1), (x2, y2), (x3, y3)]`` corresponding to the four corners. """ wd, ht = self.get_window_size() radius = int(np.ceil(math.sqrt(wd**2 + ht**2) * 0.5)) ctr_x, ctr_y = self.get_center()[:2] win_pts = np.asarray([(ctr_x - radius, ctr_y - radius), (ctr_x + radius, ctr_y - radius), (ctr_x + radius, ctr_y + radius), (ctr_x - radius, ctr_y + radius)]) arr_pts = self.tform['data_to_window'].from_(win_pts) return arr_pts
[docs] def get_datarect(self): """Get the approximate LL and UR corners of the displayed image. Returns ------- rect : tuple Bounding box in data coordinates in the form of ``(x1, y1, x2, y2)``. """ # get the data points in the four corners a, b = trcalc.get_bounds(self.get_pan_rect()) # determine bounding box x1, y1 = a[:2] x2, y2 = b[:2] return (x1, y1, x2, y2)
[docs] def get_data(self, data_x, data_y): """Get the data value at the given position. Indices are zero-based, as in Numpy. Parameters ---------- data_x, data_y : int Data indices for X and Y, respectively. Returns ------- value Data value. """ return self.vip.get_data_xy(data_x, data_y)
[docs] def get_pixel_distance(self, x1, y1, x2, y2): """Calculate distance between the given pixel positions. Parameters ---------- x1, y1, x2, y2 : number Pixel coordinates. Returns ------- dist : float Rounded distance. """ dx = abs(x2 - x1) dy = abs(y2 - y1) dist = np.sqrt(dx * dx + dy * dy) dist = np.round(dist) return dist
def _sanity_check_scale(self, scale_x, scale_y): """Do a sanity check on the proposed scale vs. window size. Raises an exception if there will be a problem. """ win_wd, win_ht = self.get_window_size() if (win_wd <= 0) or (win_ht <= 0): raise ImageViewError("window size undefined") # final sanity check on resulting output image size sx = float(win_wd) / scale_x sy = float(win_ht) / scale_y if (sx < 1.0) or (sy < 1.0): raise ValueError( "resulting scale (%f, %f) would result in pixel size " "approaching window size" % (scale_x, scale_y)) def _reset_bbox(self): """This function should only be called internally. It resets the viewers bounding box based on changes to pan or scale. """ scale_x, scale_y = self.get_scale_xy() pan_x, pan_y = self.get_pan(coord='data')[:2] win_wd, win_ht = self.get_window_size() # NOTE: need to set at least a minimum 1-pixel dimension on # the window or we get a scale calculation exception. See github # issue 431 win_wd, win_ht = max(1, win_wd), max(1, win_ht) self.renderer._confirm_pan_and_scale(scale_x, scale_y, pan_x, pan_y, win_wd, win_ht)
[docs] def set_scale(self, scale, no_reset=False): """Scale the image in a channel. Also see :meth:`zoom_to`. Parameters ---------- scale : tuple of float Scaling factors for the image in the X and Y axes. no_reset : bool Do not reset ``autozoom`` setting. """ return self.scale_to(*scale[:2], no_reset=no_reset)
[docs] def scale_to(self, scale_x, scale_y, no_reset=False): """Scale the image in a channel. This only changes the viewer settings; the image is not modified in any way. Also see :meth:`zoom_to`. Parameters ---------- scale_x, scale_y : float Scaling factors for the image in the X and Y axes, respectively. no_reset : bool Do not reset ``autozoom`` setting. """ try: self._sanity_check_scale(scale_x, scale_y) except Exception as e: self.logger.warning("Error in scaling: %s" % (str(e))) return ratio = float(scale_x) / float(scale_y) if ratio < 1.0: # Y is stretched scale_x_base, scale_y_base = 1.0, 1.0 / ratio else: # X may be stretched scale_x_base, scale_y_base = ratio, 1.0 self.t_.set(scale_x_base=scale_x_base, scale_y_base=scale_y_base) self._scale_to(scale_x, scale_y, no_reset=no_reset)
def _scale_to(self, scale_x, scale_y, no_reset=False): # Check scale limits maxscale = max(scale_x, scale_y) max_lim = self.t_.get('scale_max', None) if (max_lim is not None) and (maxscale > max_lim): self.logger.warning("Scale (%.2f) exceeds max scale limit (%.2f)" % ( maxscale, self.t_['scale_max'])) # TODO: exception? return minscale = min(scale_x, scale_y) min_lim = self.t_.get('scale_min', None) if (min_lim is not None) and (minscale < min_lim): self.logger.warning("Scale (%.2f) exceeds min scale limit (%.2f)" % ( minscale, self.t_['scale_min'])) # TODO: exception? return # Sanity check on the scale vs. window size try: self._sanity_check_scale(scale_x, scale_y) except Exception as e: self.logger.warning("Error in scaling: %s" % (str(e))) return self.t_.set(scale=(scale_x, scale_y)) self._reset_bbox() # If user specified "override" or "once" for auto zoom, then turn off # auto zoom now that they have set the zoom manually if (not no_reset) and (self.t_['autozoom'] in ('override', 'once')): self.t_.set(autozoom='off')
[docs] def scale_cb(self, setting, value): """Handle callback related to image scaling.""" zoomlevel = self.zoom.calc_level(value) self.t_.set(zoomlevel=zoomlevel) self.renderer.scale(value)
[docs] def get_scale(self): """Same as :meth:`get_scale_max`.""" return self.get_scale_max()
[docs] def get_scale_max(self): """Get maximum scale factor. Returns ------- scalefactor : float Scale factor for X or Y, whichever is larger. """ scale = self.get_scale_xy() scalefactor = max(*scale) return scalefactor
[docs] def get_scale_min(self): """Get minimum scale factor. Returns ------- scalefactor : float Scale factor for X or Y, whichever is smaller. """ scale = self.get_scale_xy() scalefactor = min(*scale) return scalefactor
[docs] def get_scale_xy(self): """Get scale factors. Returns ------- scalefactors : tuple Scale factors for X and Y, in that order. """ return self.t_['scale'][:2]
[docs] def get_scale_base_xy(self): """Get stretch factors. Returns ------- stretchfactors : tuple Stretch factors for X and Y, in that order. """ return (self.t_['scale_x_base'], self.t_['scale_y_base'])
[docs] def set_scale_base_xy(self, scale_x_base, scale_y_base): """Set stretch factors. Parameters ---------- scale_x_base, scale_y_base : float Stretch factors for X and Y, respectively. """ self.t_.set(scale_x_base=scale_x_base, scale_y_base=scale_y_base)
[docs] def get_scale_text(self): """Report current scaling in human-readable format. Returns ------- text : str ``'<num> x'`` if enlarged, or ``'1/<num> x'`` if shrunken. """ scalefactor = self.get_scale_max() if scalefactor >= 1.0: text = '%.2fx' % (scalefactor) else: text = '1/%.2fx' % (1.0 / scalefactor) return text
[docs] def zoom_to(self, zoomlevel, no_reset=False): """Set zoom level in a channel. This only changes the relevant settings; The image is not modified. Also see :meth:`scale_to`. .. note:: In addition to the given zoom level, other zoom settings are defined for the channel in preferences. Parameters ---------- zoomlevel : int The zoom level to zoom the image. Negative value to zoom out; positive to zoom in. no_reset : bool Do not reset ``autozoom`` setting. """ scale_x, scale_y = self.zoom.calc_scale(zoomlevel) self._scale_to(scale_x, scale_y, no_reset=no_reset)
[docs] def zoom_in(self, incr=1.0): """Zoom in a level. Also see :meth:`zoom_to`. Parameters ---------- incr : float (optional, defaults to 1) The value to increase the zoom level """ level = self.zoom.calc_level(self.t_['scale']) self.zoom_to(level + incr)
[docs] def zoom_out(self, decr=1.0): """Zoom out a level. Also see :meth:`zoom_to`. Parameters ---------- decr : float (optional, defaults to 1) The value to decrease the zoom level """ level = self.zoom.calc_level(self.t_['scale']) self.zoom_to(level - decr)
[docs] def zoom_fit(self, no_reset=False): """Zoom to fit display window. Also see :meth:`zoom_to`. Parameters ---------- no_reset : bool Do not reset ``autozoom`` setting. """ # calculate actual width of the limits/image, considering rotation xy_mn, xy_mx = self.get_limits() try: wwidth, wheight = self.get_window_size() self.logger.debug("Window size is %dx%d" % (wwidth, wheight)) if self.t_['swap_xy']: wwidth, wheight = wheight, wwidth except Exception: return # zoom_fit also centers image with self.suppress_redraw: self.center_image(no_reset=no_reset) ctr_x, ctr_y, rot_deg = self.get_rotation_info() # Find min/max extents of limits as shown by viewer xs = np.array((xy_mn[0], xy_mx[0], xy_mx[0], xy_mn[0])) ys = np.array((xy_mn[1], xy_mn[1], xy_mx[1], xy_mx[1])) x0, y0 = trcalc.rotate_pt(xs, ys, rot_deg, xoff=ctr_x, yoff=ctr_y) min_x, min_y = np.min(x0), np.min(y0) max_x, max_y = np.max(x0), np.max(y0) width, height = max_x - min_x, max_y - min_y if min(width, height) <= 0: return # Calculate optimum zoom level to still fit the window size scale_x = (float(wwidth) / (float(width) * self.t_['scale_x_base'])) scale_y = (float(wheight) / (float(height) * self.t_['scale_y_base'])) scalefactor = min(scale_x, scale_y) # account for t_[scale_x/y_base] scale_x = scalefactor * self.t_['scale_x_base'] scale_y = scalefactor * self.t_['scale_y_base'] self._scale_to(scale_x, scale_y, no_reset=no_reset) if self.t_['autozoom'] == 'once': self.t_.set(autozoom='off')
[docs] def get_zoom(self): """Get zoom level. Returns ------- zoomlevel : float Zoom level. """ return self.zoom.calc_level(self.t_['scale'])
[docs] def get_zoomrate(self): """Get zoom rate. Returns ------- zoomrate : float Zoom rate. """ return self.t_['zoom_rate']
[docs] def set_zoomrate(self, zoomrate): """Set zoom rate. Parameters ---------- zoomrate : float Zoom rate. """ self.t_.set(zoom_rate=zoomrate)
[docs] def get_zoom_algorithm(self): """Get zoom algorithm. Returns ------- name : str Name of the zoom algorithm in use. """ return str(self.zoom)
[docs] def set_zoom_algorithm(self, name): """Set zoom algorithm. Parameters ---------- name : str Name of a zoom algorithm to use. """ name = name.lower() alg_names = list(zoom.get_zoom_alg_names()) if name not in alg_names: raise ImageViewError("Alg '%s' must be one of: %s" % ( name, ', '.join(alg_names))) self.t_.set(zoom_algorithm=name)
[docs] def zoomsetting_change_cb(self, setting, value): """Handle callback related to changes in zoom.""" alg_name = self.t_['zoom_algorithm'] self.zoom = zoom.get_zoom_alg(alg_name)(self) self.zoom_to(self.get_zoom())
[docs] def interpolation_change_cb(self, setting, value): """Handle callback related to changes in interpolation.""" self.renderer.interpolation_change(value)
[docs] def set_name(self, name): """Set viewer name.""" self.name = name
[docs] def get_scale_limits(self): """Get scale limits. Returns ------- scale_limits : tuple Minimum and maximum scale limits, respectively. """ return (self.t_['scale_min'], self.t_['scale_max'])
[docs] def set_scale_limits(self, scale_min, scale_max): """Set scale limits. Parameters ---------- scale_min, scale_max : float Minimum and maximum scale limits, respectively. """ # TODO: force scale to within limits if already outside? self.t_.set(scale_min=scale_min, scale_max=scale_max)
[docs] def enable_autozoom(self, option): """Set ``autozoom`` behavior. Parameters ---------- option : {'on', 'override', 'once', 'off'} Option for zoom behavior. A list of acceptable options can also be obtained by :meth:`get_autozoom_options`. Raises ------ ginga.ImageView.ImageViewError Invalid option. """ option = option.lower() if option not in self.autozoom_options: raise ImageViewError("Bad autozoom option '%s': must be one of %s" % ( str(self.autozoom_options))) self.t_.set(autozoom=option)
[docs] def get_autozoom_options(self): """Get all valid ``autozoom`` options. Returns ------- autozoom_options : tuple A list of valid options. """ return self.autozoom_options
[docs] def set_pan(self, pan_x, pan_y, coord='data', no_reset=False): """Set pan position. Parameters ---------- pan_x, pan_y : float Pan positions in X and Y. coord : {'data', 'wcs'} Indicates whether the given pan positions are in data or WCS space. no_reset : bool Do not reset ``autocenter`` setting. """ pan_pos = (pan_x, pan_y) with self.suppress_redraw: self.t_.set(pan=pan_pos, pan_coord=coord) self._reset_bbox() # If user specified "override" or "once" for auto center, then turn off # auto center now that they have set the pan manually if (not no_reset) and (self.t_['autocenter'] in ('override', 'once')): self.t_.set(autocenter='off')
[docs] def pan_cb(self, setting, value): """Handle callback related to changes in pan.""" pan_x, pan_y = value[:2] self.logger.debug("pan set to %.2f,%.2f" % (pan_x, pan_y)) self.renderer.pan(value)
[docs] def get_pan(self, coord='data'): """Get pan positions. Parameters ---------- coord : {'data', 'wcs'} Indicates whether the pan positions are returned in data or WCS space. Returns ------- positions : tuple X and Y positions, in that order. """ pan_x, pan_y = self.t_['pan'][:2] if coord == 'wcs': if self.t_['pan_coord'] == 'data': image = self.get_image() if image is not None: try: return image.pixtoradec(pan_x, pan_y) except Exception as e: pass # <-- data already in coordinates form return (pan_x, pan_y) # <-- requesting data coords if self.t_['pan_coord'] == 'data': return (pan_x, pan_y) image = self.get_image() if image is not None: try: return image.radectopix(pan_x, pan_y) except Exception as e: pass return (pan_x, pan_y)
[docs] def panset_xy(self, data_x, data_y, no_reset=False): """Similar to :meth:`set_pan`, except that input pan positions are always in data space. """ pan_coord = self.t_['pan_coord'] # To center on the pixel if pan_coord == 'wcs': image = self.get_image() if image is None: return pan_x, pan_y = image.pixtoradec(data_x, data_y) else: pan_x, pan_y = data_x, data_y self.set_pan(pan_x, pan_y, coord=pan_coord, no_reset=no_reset)
[docs] def panset_pct(self, pct_x, pct_y): """Similar to :meth:`set_pan`, except that pan positions are determined by multiplying data dimensions with the given scale factors, where 1 is 100%. """ xy_mn, xy_mx = self.get_limits() data_x = (xy_mn[0] + xy_mx[0]) * pct_x data_y = (xy_mn[1] + xy_mx[1]) * pct_y self.panset_xy(data_x, data_y)
[docs] def center_image(self, no_reset=True): """Pan to the center of the image. Parameters ---------- no_reset : bool See :meth:`set_pan`. """ xy_mn, xy_mx = self.get_limits() data_x = (xy_mn[0] + xy_mx[0]) * 0.5 data_y = (xy_mn[1] + xy_mx[1]) * 0.5 self.panset_xy(data_x, data_y, no_reset=no_reset) if self.t_['autocenter'] == 'once': self.t_.set(autocenter='off')
[docs] def enable_autocenter(self, option): """Set ``autocenter`` behavior. Parameters ---------- option : {'on', 'override', 'once', 'off'} Option for auto-center behavior. A list of acceptable options can also be obtained by :meth:`get_autocenter_options`. Raises ------ ginga.ImageView.ImageViewError Invalid option. """ option = option.lower() assert(option in self.autocenter_options), \ ImageViewError("Bad autocenter option '%s': must be one of %s" % ( str(self.autocenter_options))) self.t_.set(autocenter=option)
set_autocenter = enable_autocenter
[docs] def get_autocenter_options(self): """Get all valid ``autocenter`` options. Returns ------- autocenter_options : tuple A list of valid options. """ return self.autocenter_options
[docs] def get_transforms(self): """Get transformations behavior. Returns ------- transforms : tuple Selected options for ``flip_x``, ``flip_y``, and ``swap_xy``. """ return (self.t_['flip_x'], self.t_['flip_y'], self.t_['swap_xy'])
[docs] def get_cut_levels(self): """Get cut levels. Returns ------- cuts : tuple Low and high values, in that order. """ return self.t_['cuts']
[docs] def cut_levels(self, loval, hival, no_reset=False): """Apply cut levels on the image view. Parameters ---------- loval, hival : float Low and high values of the cut levels, respectively. no_reset : bool Do not reset ``autocuts`` setting. """ self.t_.set(cuts=(loval, hival)) # If user specified "override" or "once" for auto levels, # then turn off auto levels now that they have set the levels # manually if (not no_reset) and (self.t_['autocuts'] in ('once', 'override')): self.t_.set(autocuts='off')
[docs] def auto_levels(self, autocuts=None): """Apply auto-cut levels on the image view. Parameters ---------- autocuts : subclass of `~ginga.AutoCuts.AutoCutsBase` or `None` An object that implements the desired auto-cut algorithm. If not given, use algorithm from preferences. """ if autocuts is None: autocuts = self.autocuts image = self.get_image() if image is None: #image = self.vip return loval, hival = autocuts.calc_cut_levels(image) # this will invoke cut_levels_cb() self.t_.set(cuts=(loval, hival)) # If user specified "once" for auto levels, then turn off # auto levels now that we have cut levels established if self.t_['autocuts'] == 'once': self.t_.set(autocuts='off')
[docs] def auto_levels_cb(self, setting, value): """Handle callback related to changes in auto-cut levels.""" # Did we change the method? method = self.t_['autocut_method'] params = self.t_.get('autocut_params', []) params = dict(params) if method != str(self.autocuts): ac_class = AutoCuts.get_autocuts(method) self.autocuts = ac_class(self.logger, **params) else: self.autocuts.update_params(**params) # Redo the auto levels #if self.t_['autocuts'] != 'off': # NOTE: users seem to expect that when the auto cuts parameters # are changed that the cuts should be immediately recalculated self.auto_levels()
[docs] def cut_levels_cb(self, setting, value): """Handle callback related to changes in cut levels.""" self.renderer.levels_change(value)
[docs] def enable_autocuts(self, option): """Set ``autocuts`` behavior. Parameters ---------- option : {'on', 'override', 'once', 'off'} Option for auto-cut behavior. A list of acceptable options can also be obtained by :meth:`get_autocuts_options`. Raises ------ ginga.ImageView.ImageViewError Invalid option. """ option = option.lower() assert(option in self.autocuts_options), \ ImageViewError("Bad autocuts option '%s': must be one of %s" % ( str(self.autocuts_options))) self.t_.set(autocuts=option)
[docs] def get_autocuts_options(self): """Get all valid ``autocuts`` options. Returns ------- autocuts_options : tuple A list of valid options. """ return self.autocuts_options
[docs] def set_autocut_params(self, method, **params): """Set auto-cut parameters. Parameters ---------- method : str Auto-cut algorithm. A list of acceptable options can be obtained by :meth:`get_autocut_methods`. params : dict Algorithm-specific keywords and values. """ self.logger.debug("Setting autocut params method=%s params=%s" % ( method, str(params))) params = list(params.items()) self.t_.set(autocut_method=method, autocut_params=params)
[docs] def get_autocut_methods(self): """Same as :meth:`ginga.AutoCuts.AutoCutsBase.get_algorithms`.""" return self.autocuts.get_algorithms()
[docs] def set_autocuts(self, autocuts): """Set the auto-cut algorithm. Parameters ---------- autocuts : subclass of `~ginga.AutoCuts.AutoCutsBase` An object that implements the desired auto-cut algorithm. """ self.autocuts = autocuts
[docs] def transform(self, flip_x, flip_y, swap_xy): """Transform view of the image. .. note:: Transforming the image is generally faster than rotating, if rotating in 90 degree increments. Also see :meth:`rotate`. Parameters ---------- flipx, flipy : bool If `True`, flip the image in the X and Y axes, respectively swapxy : bool If `True`, swap the X and Y axes. """ self.logger.debug("flip_x=%s flip_y=%s swap_xy=%s" % ( flip_x, flip_y, swap_xy)) with self.suppress_redraw: self.t_.set(flip_x=flip_x, flip_y=flip_y, swap_xy=swap_xy)
[docs] def transform_cb(self, setting, value): """Handle callback related to changes in transformations.""" self.make_callback('transform') state = (self.t_['flip_x'], self.t_['flip_y'], self.t_['swap_xy']) self.renderer.transform_2d(state)
[docs] def copy_attributes(self, dst_fi, attrlist, share=False, whence=0): """Copy interesting attributes of our configuration to another image view. Parameters ---------- dst_fi : subclass of `ImageViewBase` Another instance of image view. attrlist : list A list of attribute names to copy. They can be ``'transforms'``, ``'rotation'``, ``'cutlevels'``, ``'rgbmap'``, ``'zoom'``, ``'pan'``, ``'autocuts'``, ``'limits'``, ``'icc'`` or ``'interpolation'``. share : bool If True, the designated settings will be shared, otherwise the values are simply copied. """ # TODO: change API to just go with settings names? keylist = [] _whence = 3.0 if whence <= 0.0: if 'limits' in attrlist: keylist.extend(['limits']) _whence = min(_whence, 0.0) if 'zoom' in attrlist: keylist.extend(['scale']) _whence = min(_whence, 0.0) if 'interpolation' in attrlist: keylist.extend(['interpolation']) _whence = min(_whence, 0.0) if 'pan' in attrlist: keylist.extend(['pan']) _whence = min(_whence, 0.0) if whence <= 1.0: if 'autocuts' in attrlist: keylist.extend(['autocut_method', 'autocut_params']) _whence = min(_whence, 1.0) if 'cutlevels' in attrlist: keylist.extend(['cuts']) _whence = min(_whence, 1.0) if whence <= 2.0: if 'rgbmap' in attrlist: keylist.extend(['color_algorithm', 'color_hashsize', 'color_map', 'intensity_map', 'color_array', 'shift_array']) _whence = min(_whence, 2.0) if whence <= 2.3: if 'icc' in attrlist: keylist.extend(['icc_output_profile', 'icc_output_intent', 'icc_proof_profile', 'icc_proof_intent', 'icc_black_point_compensation']) _whence = min(_whence, 2.3) if whence <= 2.5: if 'transforms' in attrlist: keylist.extend(['flip_x', 'flip_y', 'swap_xy']) _whence = min(_whence, 2.5) if whence <= 2.6: if 'rotation' in attrlist: keylist.extend(['rot_deg']) _whence = min(_whence, 2.6) whence = max(_whence, whence) with dst_fi.suppress_redraw: if share: self.t_.share_settings(dst_fi.get_settings(), keylist=keylist) else: self.t_.copy_settings(dst_fi.get_settings(), keylist=keylist)
[docs] def get_rotation(self): """Get image rotation angle. Returns ------- rot_deg : float Rotation angle in degrees. """ return self.t_['rot_deg']
[docs] def rotate(self, deg): """Rotate the view of an image in a channel. .. note:: Transforming the image is generally faster than rotating, if rotating in 90 degree increments. Also see :meth:`transform`. Parameters ---------- deg : float Rotation angle in degrees. """ self.t_.set(rot_deg=deg)
[docs] def rotation_change_cb(self, setting, value): """Handle callback related to changes in rotation angle.""" self.renderer.rotate_2d(value)
[docs] def get_center(self): """Get image center. Returns ------- ctr : tuple X and Y positions, in that order. """ center = self.renderer.get_center()[:2] return center
[docs] def get_rgb_order(self): """Get RGB order. Returns ------- rgb : str Returns the order of RGBA planes required by the subclass to render the canvas properly. """ return self.rgb_order
[docs] def get_rotation_info(self): """Get rotation information. Returns ------- info : tuple X and Y positions, and rotation angle in degrees, in that order. """ win_x, win_y = self.get_center() return (win_x, win_y, self.t_['rot_deg'])
[docs] def enable_auto_orient(self, tf): """Set ``auto_orient`` behavior. Parameters ---------- tf : bool Turns automatic image orientation on or off. """ self.t_.set(auto_orient=tf)
[docs] def auto_orient(self): """Set the orientation for the image to a reasonable default.""" image = self.get_image() if image is None: return invert_y = not isinstance(image, AstroImage.AstroImage) # Check for various things to set based on metadata header = image.get_header() if header: # Auto-orientation orient = header.get('Orientation', None) if orient is None: orient = header.get('Image Orientation', None) if orient is not None: self.logger.debug("orientation [%s]" % orient) try: orient = int(str(orient)) self.logger.info( "setting orientation from metadata [%d]" % (orient)) flip_x, flip_y, swap_xy = self.orient_map[orient] self.transform(flip_x, flip_y, swap_xy) invert_y = False except Exception as e: # problems figuring out orientation--let it be self.logger.error("orientation error: %s" % str(e)) if invert_y: flip_x, flip_y, swap_xy = self.get_transforms() #flip_y = not flip_y flip_y = True self.transform(flip_x, flip_y, swap_xy)
[docs] def get_coordmap(self, key): """Get coordinate mapper. Parameters ---------- key : str Name of the desired coordinate mapper. Returns ------- mapper Coordinate mapper object (see `ginga.canvas.coordmap`). """ return self.coordmap[key]
[docs] def set_coordmap(self, key, mapper): """Set coordinate mapper. Parameters ---------- key : str Name of the coordinate mapper. mapper Coordinate mapper object (see `ginga.canvas.coordmap`). """ self.coordmap[key] = mapper
[docs] def recalc_transforms(self, trcat=None): """Takes a catalog of transforms (`trcat`) and builds the chain of default transforms necessary to do rendering with most backends. """ if trcat is None: trcat = self.trcat self.tform = { 'window_to_native': trcat.WindowNativeTransform(self), 'cartesian_to_window': trcat.CartesianWindowTransform(self), 'cartesian_to_native': (trcat.RotationFlipTransform(self) + trcat.CartesianNativeTransform(self)), 'data_to_cartesian': (trcat.DataCartesianTransform(self) + trcat.ScaleTransform(self)), 'data_to_scrollbar': (trcat.DataCartesianTransform(self) + trcat.RotationFlipTransform(self)), 'mouse_to_data': ( trcat.InvertedTransform(trcat.DataCartesianTransform(self) + trcat.ScaleTransform(self) + trcat.RotationFlipTransform(self) + trcat.CartesianNativeTransform(self))), 'data_to_window': (trcat.DataCartesianTransform(self) + trcat.ScaleTransform(self) + trcat.RotationFlipTransform(self) + trcat.CartesianWindowTransform(self)), 'data_to_percentage': (trcat.DataCartesianTransform(self) + trcat.ScaleTransform(self) + trcat.RotationFlipTransform(self) + trcat.CartesianWindowTransform(self) + trcat.WindowPercentageTransform(self)), 'data_to_native': (trcat.DataCartesianTransform(self) + trcat.ScaleTransform(self) + trcat.RotationFlipTransform(self) + trcat.CartesianNativeTransform(self)), 'wcs_to_data': trcat.WCSDataTransform(self), 'wcs_to_native': (trcat.WCSDataTransform(self) + trcat.DataCartesianTransform(self) + trcat.ScaleTransform(self) + trcat.RotationFlipTransform(self) + trcat.CartesianNativeTransform(self)), }
[docs] def set_bg(self, r, g, b): """Set the background color. Parameters ---------- r, g, b : float RGB values, which should be between 0 and 1, inclusive. """ self.set_background((r, g, b))
[docs] def set_background(self, bg): """Set the background color. Parameters ---------- bg : str or tuple of float color name or tuple of floats, between 0 and 1, inclusive. """ self.img_bg = colors.resolve_color(bg) self.renderer.bg_change(self.img_bg)
[docs] def get_bg(self): """Get the background color. Returns ------- img_bg : tuple RGB values. """ return self.img_bg
[docs] def set_fg(self, r, g, b): """Set the foreground color. Parameters ---------- r, g, b : float RGB values, which should be between 0 and 1, inclusive. """ self.set_foreground((r, g, b))
[docs] def set_foreground(self, fg): """Set the foreground color. Parameters ---------- fg : str or tuple of float color name or tuple of floats, between 0 and 1, inclusive. """ self.img_fg = colors.resolve_color(fg) self.renderer.fg_change(self.img_fg)
[docs] def get_fg(self): """Get the foreground color. Returns ------- img_fg : tuple RGB values. """ return self.img_fg
[docs] def is_compound(self): """Indicate if canvas object is a compound object. This can be re-implemented by subclasses that can overplot objects. Returns ------- status : bool Currently, this *always* returns `False`. """ return False
[docs] def window_has_origin_upper(self): """Indicate if window of backend toolkit is implemented with an origin up or down. Returns ------- res : bool Returns `True` if the origin is up, `False` otherwise. """ return self.origin_upper
[docs] def get_last_win_xy(self): """Get the last position of the cursor in window coordinates. This can be overridden by subclasses, if necessary. """ return (self.last_win_x, self.last_win_y)
[docs] def get_last_data_xy(self): """Get the last position of the cursor in data coordinates. This can be overridden by subclasses, if necessary. """ return (self.last_data_x, self.last_data_y)
[docs] def onscreen_message(self, text, delay=None, redraw=True): """Place a message onscreen in the viewer window. This must be implemented by subclasses. Parameters ---------- text : str the text to draw in the window delay : float or None if None, the message will remain until another message is set. If a float, specifies the time in seconds before the message will be erased. (default: None) redraw : bool True if the widget should be redrawn right away (so that the message appears). (default: True) """ self.logger.warning("Subclass should override this abstract method!")
[docs] def onscreen_message_off(self): """Erase any message onscreen in the viewer window. """ return self.onscreen_message(None)
[docs] def set_enter_focus(self, tf): """Determine whether the viewer widget should take focus when the cursor enters the window. Parameters ---------- tf : bool If True the widget will grab focus when the cursor moves into the window. """ self.t_.set(enter_focus=tf)
[docs] def update_widget(self): """Update the area corresponding to the backend widget. This must be implemented by subclasses. """ self.logger.warning("Subclass should override this abstract method!")
# TO BE DEPRECATED--please use update_widget
[docs] def update_image(self): return self.update_widget()
[docs] def reschedule_redraw(self, time_sec): """Reschedule redraw event. This should be implemented by subclasses. Parameters ---------- time_sec : float Time, in seconds, to wait. """ # subclass implements this method to call delayed_redraw() after # time_sec. If subclass does not override, redraw is immediate. self.delayed_redraw()
[docs] def set_cursor(self, cursor): """Set the cursor in the viewer widget. This should be implemented by subclasses. Parameters ---------- cursor : object a cursor object in the back end's toolkit """ self.logger.warning("Subclass should override this abstract method!")
[docs] def make_timer(self): """Return a timer object implemented using the back end. This should be implemented by subclasses. Returns ------- timer : a Timer object """ #self.logger.warning("Subclass should override this abstract method!") return None
[docs] def make_cursor(self, iconpath, x, y): """Make a cursor in the viewer's native widget toolkit. This should be implemented by subclasses. Parameters ---------- iconpath : str the path to a PNG image file defining the cursor x : int the X position of the center of the cursor hot spot y : int the Y position of the center of the cursor hot spot """ self.logger.warning("Subclass should override this abstract method!") return None
[docs] def center_cursor(self): """Center the cursor in the viewer's widget, in both X and Y. This should be implemented by subclasses. """ self.logger.warning("Subclass should override this abstract method!")
[docs] def position_cursor(self, data_x, data_y): """Position the current cursor to a location defined it data coords. This should be implemented by subclasses. Parameters ---------- data_x : float the X position to position the cursor in data coords data_y : float the X position to position the cursor in data coords """ self.logger.warning("Subclass should override this abstract method!")
[docs] def get_cursor(self, cname): """Get the cursor stored under the name. This can be overridden by subclasses, if necessary. Parameters ---------- cname : str name of the cursor to return. """ return self.cursor[cname]
[docs] def define_cursor(self, cname, cursor): """Define a viewer cursor under a name. Does not change the current cursor. Parameters ---------- cname : str name of the cursor to define. cursor : object a cursor object in the back end's toolkit `cursor` is usually constructed from `make_cursor`. """ self.cursor[cname] = cursor
[docs] def switch_cursor(self, cname): """Switch the viewer's cursor to the one defined under a name. Parameters ---------- cname : str name of the cursor to switch to. """ self.set_cursor(self.cursor[cname])
[docs] def prepare_image(self, cvs_img, cache, whence): """This can be overridden by subclasses. """ self.renderer.prepare_image(cvs_img, cache, whence)
[docs] def get_image_as_array(self, order=None): """Get the current image shown in the viewer, with any overlaid graphics, in a numpy array with channels as needed and ordered. This can be overridden by subclasses. """ if order is None: order = self.rgb_order return self.renderer.get_surface_as_array(order=order)
[docs] def get_image_as_buffer(self, output=None, order=None): """Get the current image shown in the viewer, with any overlaid graphics, in a IO buffer with channels as needed and ordered by the back end widget. This can be overridden by subclasses. Parameters ---------- output : a file IO-like object or None open python IO descriptor or None to have one created Returns ------- buffer : file IO-like object This will be the one passed in, unless `output` is None in which case a BytesIO obejct is returned """ obuf = output if obuf is None: obuf = BytesIO() arr8 = self.get_image_as_array(order=order) if not hasattr(arr8, 'tobytes'): # older versions of numpy obuf.write(arr8.tostring(order='C')) else: obuf.write(arr8.tobytes(order='C')) ## if output is not None: ## return None return obuf
[docs] def get_rgb_image_as_buffer(self, output=None, format='png', quality=90): """Get the current image shown in the viewer, with any overlaid graphics, in a file IO-like object encoded as a bitmap graphics file. This can be overridden by subclasses. Parameters ---------- output : a file IO-like object or None open python IO descriptor or None to have one created format : str A string defining the format to save the image. Typically at least 'jpeg' and 'png' are supported. (default: 'png') quality: int The quality metric for saving lossy compressed formats. Returns ------- buffer : file IO-like object This will be the one passed in, unless `output` is None in which case a BytesIO obejct is returned """ return self.renderer.get_surface_as_rgb_format_buffer( output=output, format=format, quality=quality)
[docs] def get_rgb_image_as_bytes(self, format='png', quality=90): """Get the current image shown in the viewer, with any overlaid graphics, in the form of a buffer in the form of bytes. Parameters ---------- format : str See :meth:`get_rgb_image_as_buffer`. quality: int See :meth:`get_rgb_image_as_buffer`. Returns ------- buffer : bytes The window contents as a buffer in the form of bytes. """ obuf = self.get_rgb_image_as_buffer(format=format, quality=quality) return bytes(obuf.getvalue())
[docs] def get_rgb_image_as_widget(self, output=None, format='png', quality=90): """Get the current image shown in the viewer, with any overlaid graphics, in the form of a image widget in the toolkit of the back end. Parameters ---------- See :meth:`get_rgb_image_as_buffer`. Returns ------- widget : object An image widget object in the viewer's back end toolkit """ raise ImageViewError("Subclass should override this abstract method!")
[docs] def save_rgb_image_as_file(self, filepath, format='png', quality=90): """Save the current image shown in the viewer, with any overlaid graphics, in a file with the specified format and quality. This can be overridden by subclasses. Parameters ---------- filepath : str path of the file to write format : str See :meth:`get_rgb_image_as_buffer`. quality: int See :meth:`get_rgb_image_as_buffer`. """ with open(filepath, 'wb') as out_f: self.get_rgb_image_as_buffer(output=out_f, format=format, quality=quality) self.logger.debug("wrote %s file '%s'" % (format, filepath))
[docs] def get_plain_image_as_widget(self): """Get the current image shown in the viewer, without any overlaid graphics, in the format of an image widget in the back end toolkit. Typically used for generating thumbnails. This should be implemented by subclasses. Returns ------- widget : object An image widget object in the viewer's back end toolkit """ raise ImageViewError("Subclass should override this abstract method!")
[docs] def save_plain_image_as_file(self, filepath, format='png', quality=90): """Save the current image shown in the viewer, without any overlaid graphics, in a file with the specified format and quality. Typically used for generating thumbnails. This should be implemented by subclasses. Parameters ---------- filepath : str path of the file to write format : str See :meth:`get_rgb_image_as_buffer`. quality: int See :meth:`get_rgb_image_as_buffer`. """ raise ImageViewError("Subclass should override this abstract method!")
[docs] def take_focus(self): """Have the widget associated with this viewer take the keyboard focus. This should be implemented by subclasses, if they have a widget that can take focus. """ pass
[docs] def set_onscreen_message(self, text, redraw=True): """Called by a subclass to update the onscreen message. Parameters ---------- text : str The text to show in the display. """ width, height = self.get_window_size() font = self.t_.get('onscreen_font', 'sans serif') font_size = self.t_.get('onscreen_font_size', None) if font_size is None: font_size = self._calc_font_size(width) # TODO: need some way to accurately estimate text extents # without actually putting text on the canvas ht, wd = font_size, font_size if text is not None: wd = len(text) * font_size * 1.1 x = (width // 2) - (wd // 2) y = ((height // 3) * 2) - (ht // 2) tag = '_$onscreen_msg' canvas = self.get_private_canvas() try: message = canvas.get_object_by_tag(tag) if text is None: message.text = '' else: message.x = x message.y = y message.text = text message.fontsize = font_size except KeyError: if text is None: text = '' Text = canvas.get_draw_class('text') canvas.add(Text(x, y, text=text, font=font, fontsize=font_size, color=self.img_fg, coord='window'), tag=tag, redraw=False) if redraw: canvas.update_canvas(whence=3)
def _calc_font_size(self, win_wd): """Heuristic to calculate an appropriate font size based on the width of the viewer window. Parameters ---------- win_wd : int The width of the viewer window. Returns ------- font_size : int Approximately appropriate font size in points """ font_size = 4 if win_wd >= 1600: font_size = 24 elif win_wd >= 1000: font_size = 18 elif win_wd >= 800: font_size = 16 elif win_wd >= 600: font_size = 14 elif win_wd >= 500: font_size = 12 elif win_wd >= 400: font_size = 11 elif win_wd >= 300: font_size = 10 elif win_wd >= 250: font_size = 8 elif win_wd >= 200: font_size = 6 return font_size
[docs] def show_pan_mark(self, tf, color='red'): # TO BE DEPRECATED--please use addons.show_pan_mark addons.show_pan_mark(self, tf, color=color)
[docs] def show_mode_indicator(self, tf, corner='ur'): # TO BE DEPRECATED--please use addons.show_mode_indicator addons.show_mode_indicator(self, tf, corner=corner)
[docs] def show_color_bar(self, tf, side='bottom'): # TO BE DEPRECATED--please use addons.show_color_bar addons.show_color_bar(self, tf, side=side)
[docs] def show_focus_indicator(self, tf, color='white'): # TO BE DEPRECATED--please use addons.show_focus_indicator addons.show_focus_indicator(self, tf, color=color)
class SuppressRedraw(object): def __init__(self, viewer): self.viewer = viewer def __enter__(self): self.viewer._hold_redraw_cnt += 1 return self def __exit__(self, exc_type, exc_val, exc_tb): self.viewer._hold_redraw_cnt -= 1 if (self.viewer._hold_redraw_cnt <= 0): # whence should be largest possible whence = self.viewer._defer_whence self.viewer.redraw(whence=whence) return False # END