Matplotlib was designed in terms of DPI (Dots Per Inch) which is somewhat consistent but not always exact.

If one wants an image with an exact width and height (in pixels), tricks can be used in matplotlib to (usually) approximate the desired width and height.

Whenever faced with getting exact placement of graphics and image sizes, I ended up creating whatever graphics functionality I needed as a class and then use the canvas from the PIL (Python Imaging Library) rather than use matplotlib - other than for charts and graphs.

To get close to the desired width and height of an image, one can play "tricks" with the DPI parameter of matplotlib.

The following code shows this by example. This also shows how to save multiple images of the same figure and the method can be used to save multiple formats of the same figure.

The first example program uses explicit local/global variables and calls.

The second program abstracts the first example program into a class called plot.

To third program abstracts the essential code into the module rsPlot containing class plot.

The module rsPlot and class plot will be used in most matplotlib examples.

Here is the Python code [#1]
import os import matplotlib as mpl import matplotlib.pyplot as plt from PIL import Image imagePos1 = 0 rlen1 = 1 clen1 = 1 dpi1 = 72 width1 = 640 height1 = 480 figSize1 = (width1 / dpi1, height1/dpi1) fig1, ax1 = plt.subplots(rlen1,clen1,figsize=figSize1,dpi=dpi1) print("Desired image size: {0:d} x {1:d} at {2:d} DPI".format(width1,height1,dpi1)) print("Default:") imagePos1 += 1 saveFs1 = "./color10-1-0{0:d}.png".format(imagePos1) fig1.savefig(saveFs1) savedImage1 = Image.open(saveFs1) savedWidth1, savedHeight1= savedImage1.size (_, _, _, _, _, _, savedSize1, _, savedTime1, _) = os.stat(saveFs1) print("\t{0:s} : {1:d} x {2:d} , {3:,} bytes".format(saveFs1,savedWidth1,savedHeight1,savedSize1)) print("Tight:") fig1.tight_layout() imagePos1 += 1 saveFs1 = "./color10-1-0{0:d}.png".format(imagePos1) fig1.savefig(saveFs1,bbox_inches="tight") savedImage1 = Image.open(saveFs1) savedWidth1, savedHeight1= savedImage1.size (_, _, _, _, _, _, savedSize1, _, savedTime1, _) = os.stat(saveFs1) print("\t{0:s} : {1:d} x {2:d} , {3:,} bytes".format(saveFs1,savedWidth1,savedHeight1,savedSize1))

Here is the output of the Python code.
Desired image size: 640 x 480 at 72 DPI Default: ./color10-1-01.png : 640 x 480 , 5,879 bytes Tight: ./color10-1-02.png : 632 x 472 , 5,822 bytes

The first image does have the desired width and height but has extra white space around the edge.

Note: The results can vary a little if the DPI is not an even multiple (i.e., other then base 2) of the desired width and height.

The tight_layout function adds extra space between subplots (not relevant here since there is one row and one column).

The bbox_inches parameter removes extra space at the outside of the image.

As a result, the second image does not have the desired width and height but does not have the extra white space either.

Here is the Python code [#2]
import os import matplotlib as mpl import matplotlib.pyplot as plt from PIL import Image fontDict1 = { "family" : "Verdana", "weight" : "bold", "size" : 20 } plt.rc("font", **fontDict1) mpl.rcParams["ytick.major.pad"] = "8" mpl.rcParams["xtick.major.pad"] = "8" imagePos1 = 0 basePs1 = "./" class plot(): def __init__(self, stub1, rlen1=1, clen1=1, dpi1=72, width1=640, height1=480): self.stub1 = stub1 self.rlen1 = rlen1 self.clen1 = clen1 self.dpi1 = dpi1 self.width1 = width1 self.height1 = height1 self.figSize1 = (self.width1 / self.dpi1, self.height1/self.dpi1) self.fig1, self.ax1 = plt.subplots(self.rlen1,self.clen1,figsize=self.figSize1,dpi=self.dpi1) def save(self, type1="png", tight1="tight"): global imagePos1 imagePos1 += 1 self.type1 = type1 self.tight1 = tight1 self.saveFn1 = "{0:s}-0{1:d}".format(self.stub1,imagePos1) self.saveFt1 = "{0:s}".format(self.type1) self.saveFs1 = "{0:s}{1:s}.{2:s}".format(basePs1,self.saveFn1,self.saveFt1) self.fig1.savefig(self.saveFs1,bbox_inches=tight1) self.savedImage1 = Image.open(self.saveFs1) self.savedWidth1, self.savedHeight1= self.savedImage1.size (_, _, _, _, _, _, self.savedSize1, _, self.savedTime1, _) = os.stat(self.saveFs1) print("\t{0:s} : {1:d} x {2:d} , {3:,} bytes".format(self.saveFs1,self.savedWidth1,self.savedHeight1,self.savedSize1)) plot1 = plot("color10-2") print("Desired image size: {0:d} x {1:d} at {2:d} DPI".format(plot1.width1,plot1.height1,plot1.dpi1)) print("Default:") plot1.save(tight1=None) print("Tight:") plot1.save()
Note that this code has some default font settings - not in the original code. So the border effect is similar but the actual text on the images is different.

The code shows the way in which (global) matplotlib settings can be changed. There are ways to change local settings.

Here is the output of the Python code.
Desired image size: 640 x 480 at 72 DPI Default: ./color10-2-01.png : 640 x 480 , 10,510 bytes Tight: ./color10-2-02.png : 575 x 421 , 9,977 bytes

The same image results (desired width and height, but extra white space around the edges, border shown for comparison).

The same image results (reduced width and height, but removed white space around the edges, border show for comparison).

Here is a code module in Python [module rsPlot in rsPlot.py].
# module rsPlot.py import os import matplotlib as mpl import matplotlib.pyplot as plt from PIL import Image fontDict1 = { "family" : "Verdana", "weight" : "bold", "size" : 20 } plt.rc("font", **fontDict1) mpl.rcParams["ytick.major.pad"] = "8" mpl.rcParams["xtick.major.pad"] = "8" imagePos1 = 0 basePs1 = "./" class plot(): def __init__(self, stub1, rlen1=1, clen1=1, dpi1=72, width1=640, height1=480): self.stub1 = stub1 self.rlen1 = rlen1 self.clen1 = clen1 self.dpi1 = dpi1 self.width1 = width1 self.height1 = height1 self.figSize1 = (self.width1 / self.dpi1, self.height1/self.dpi1) self.fig1, self.ax1 = plt.subplots(self.rlen1,self.clen1,figsize=self.figSize1,dpi=self.dpi1) def save(self, type1="png", tight1="tight"): global imagePos1 imagePos1 += 1 self.type1 = type1 self.tight1 = tight1 self.saveFn1 = "{0:s}-0{1:d}".format(self.stub1,imagePos1) self.saveFt1 = "{0:s}".format(self.type1) self.saveFs1 = "{0:s}{1:s}.{2:s}".format(basePs1,self.saveFn1,self.saveFt1) self.fig1.savefig(self.saveFs1,bbox_inches=tight1) self.savedImage1 = Image.open(self.saveFs1) self.savedWidth1, self.savedHeight1= self.savedImage1.size (_, _, _, _, _, _, self.savedSize1, _, self.savedTime1, _) = os.stat(self.saveFs1) print("\t{0:s} : {1:d} x {2:d} , {3:,} bytes".format(self.saveFs1,self.savedWidth1,self.savedHeight1,self.savedSize1)) print("(module rsPlot imported)")
Note that the module code has the default font settings - not in the original code.

Here is the Python code [#4]
import matplotlib.pyplot as plt import rsPlot plot1 = rsPlot.plot("color10-4") print("Desired image size: {0:d} x {1:d} at {2:d} DPI".format(plot1.width1,plot1.height1,plot1.dpi1)) print("Default:") plot1.save(tight1=None) print("Tight:") plot1.save()
Note that the same images result.

Here is the output of the Python code.
(module rsPlot imported) Desired image size: 640 x 480 at 72 DPI Default: ./color10-4-01.png : 640 x 480 , 10,510 bytes Tight: ./color10-4-02.png : 575 x 421 , 9,977 bytes

The same image results (desired width and height, but extra white space around the edges, border shown for comparison).

The same image results (reduced width and height, but removed white space around the edges, border show for comparison).