Plot multiple images with identical color assignments using matplotlib's imshow

Question

I have multiple images (numpy arrays) whose data values correspond to N different classes. Each image does not necessarily contain examples of each class. For example, there might be a total of 12 different classes (0:11), however, one image might only contain classes 1:9.

I would like to plot each image such that the color assigned to each class is the same across all images.

I've looked into several answers: here the accepted and popular answers didn't work across multiple images. here seems like it could work but I would really like to use a color map (from matplotlib import cm) so as not to manually set colors. I would also like a means to create an appropriate colorbar containing all classes.

The code I've tried is below:

import numpy as np
from matplotlib import cm
import matplotlib.pyplot as plt

t1 = np.arange(9).reshape(3,3)
t2 = t1.copy()
t2[1,1] = 10
t3 = t2.copy()
t3[1,1] = 11

cmap = cm.get_cmap('tab20', 11)

fig, axs = plt.subplots(1,3)

axs[0].imshow(t1, cmap = cmap, vmin = 0, vmax = 11)
axs[1].imshow(t2, cmap = cmap, vmin = 0, vmax = 11)
axs[2].imshow(t3, cmap = cmap, vmin = 0, vmax = 11)

`cm.get_cmap('tab20', 11)` only contains 11 different colors, while you need 12 — JohanC, Apr 22 '21 at 23:35

score 2 · Accepted Answer · answered Apr 22 '21 at 23:43

For future reference, in case you want to define your own colors and not a predefined cmap, I created the following code specifically for this some time ago.

import matplotlib as mpl
from matplotlib import cm
import matplotlib.pyplot as plt
import numpy as np

C_p = 11 # Classes

colour_names = [ # Your predefined colours
    "blue",
    "red",
    "yellow",
    "orange",
    "black",
    "purple",
    "green",
    "turquoise",
    "grey",
    "maroon",
    "silver",
    "white"
]

colour_dict = { # Color mapping (class -> colour)
    i: mpl.colors.to_rgb(colour_names[i])
    for i in range(C_p + 1)
}

# Create a colormap (optional)
colours_rgb = [colour_dict[i] for i in range(C_p)]
colours = mpl.colors.ListedColormap(colours_rgb)

norm = mpl.colors.BoundaryNorm(np.arange(C_p + 1) - 0.5, C_p)

plt.figure() # If you only want to plot one
plt.imshow(t2, cmap=colours, norm=norm)
cb = plt.colorbar(ticks=np.arange(C_p))
plt.axis("off")

Example with your t1, t2 and t3:

fig, axs = plt.subplots(1,3)
axs[0].imshow(t1, cmap = colours, norm=norm)
axs[0].set_title("t1")
axs[0].axis('off')
axs[1].imshow(t2, cmap = colours, norm=norm)
axs[1].set_title("t2")
axs[1].axis('off')
im = axs[2].imshow(t3, cmap = colours, norm=norm)
axs[2].set_title("t3")
axs[2].axis('off')
p0 = axs[0].get_position().get_points().flatten()
p1 = axs[1].get_position().get_points().flatten()
p2 = axs[2].get_position().get_points().flatten()
ax_cbar = fig.add_axes([p0[0], 0.08, p2[0], 0.05])
plt.colorbar(im, cax=ax_cbar, ticks=np.arange(C_p), orientation='horizontal')
fig.tight_layout()

Note that simply `mpl.colors.ListedColormap(colour_names)` would also work. — JohanC, Apr 22 '21 at 23:52
Oh, I hadn't noticed that. I thought it needed rgb values. Thanks — Camilo Martinez M., Apr 23 '21 at 00:16
This is the more complete solution given it has a means to create a colorbar. — Alex Witsil, Apr 24 '21 at 17:53

score 0 · Answer 2 · answered Apr 22 '21 at 23:19

Looks like cm.get_cmap needs to be adjusted to handle all the possible categories/classes in the images. The below code works:

import numpy as np
from matplotlib import cm
import matplotlib.pyplot as plt

t1 = np.arange(9).reshape(3,3)
t2 = t1.copy()
t2[1,1] = 10
t3 = t2.copy()
t3[1,1] = 11

cmap = cm.get_cmap('tab20', 12)

fig, axs = plt.subplots(1,3)

axs[0].imshow(t1, cmap = cmap, vmin = 0, vmax = 11)
axs[1].imshow(t2, cmap = cmap, vmin = 0, vmax = 11)
axs[2].imshow(t3, cmap = cmap, vmin = 0, vmax = 11)

Plot multiple images with identical color assignments using matplotlib's imshow

2 Answers2