I'm using matplotlib to plot log-normalized images but I would like the original raw image data to be represented in the colorbar rather than the [0-1] interval. I get the feeling there's a more matplotlib'y way of doing this by using some sort of normalization object and not transforming the data beforehand... in any case, there could be negative values in the raw image.
import matplotlib.pyplot as plt import numpy as np def log_transform(im): '''returns log(image) scaled to the interval [0,1]''' try: (min, max) = (im[im > 0].min(), im.max()) if (max > min) and (max > 0): return (np.log(im.clip(min, max)) - np.log(min)) / (np.log(max) - np.log(min)) except: pass return im a = np.ones((100,100)) for i in range(100): a[i] = i f = plt.figure() ax = f.add_subplot(111) res = ax.imshow(log_transform(a)) # the colorbar drawn shows [0-1], but I want to see [0-99] cb = f.colorbar(res)
I've tried using cb.set_array, but that didn't appear to do anything, and cb.set_clim, but that rescales the colors completely.
Yes, there is! Use
LogNorm. Here is a code excerpt from a utility that I wrote to display confusion matrices on a log scale.
from pylab import figure, cm from matplotlib.colors import LogNorm # C = some matrix f = figure(figsize=(6.2,5.6)) ax = f.add_axes([0.17, 0.02, 0.72, 0.79]) axcolor = f.add_axes([0.90, 0.02, 0.03, 0.79]) im = ax.matshow(C, cmap=cm.gray_r, norm=LogNorm(vmin=0.01, vmax=1)) t = [0.01, 0.1, 0.2, 0.4, 0.6, 0.8, 1.0] f.colorbar(im, cax=axcolor, ticks=t, format='$%.2f$') f.show()