savefig without frames, axes, only content

Question

In numpy/scipy I have an image stored in an array. I can display it, I want to save it using savefig without any borders, axes, labels, titles,... Just pure image, nothing else.

I want to avoid packages like PyPNG or scipy.misc.imsave, they are sometimes problematic (they do not always install well, only basic savefig() for me

Salk

Answer 1

EDIT

Changed aspect='normal to aspect='auto' since that changed in more recent versions of matplotlib (thanks to @Luke19).

---

Assuming :

import matplotlib.pyplot as plt

To make a figure without the frame :

fig = plt.figure(frameon=False)
fig.set_size_inches(w,h)

To make the content fill the whole figure

ax = plt.Axes(fig, [0., 0., 1., 1.])
ax.set_axis_off()
fig.add_axes(ax)

Then draw your image on it :

ax.imshow(your_image, aspect='auto')
fig.savefig(fname, dpi)

The aspect parameter changes the pixel size to make sure they fill the figure size specified in fig.set_size_inches(…). To get a feel of how to play with this sort of things, read through matplotlib's documentation, particularly on the subject of Axes, Axis and Artist.

Downpipe

nope, I still have some small transparent border, and what I want is no border at all, pure image – Salk

Maybe, it's because I forgot about the case where the image is not square :P. Just edited to add the aspect param. How's it now? – Downpipe

grrr, no, still the same. There is a small, transparent border around the image, few pixels on each side – Salk

almost! I figured out that the border was added by bbox_inches='tight'. But, now the images instead of desired 24x24px are 800x600px. Still looking... – Salk

If you manually set the w and h parameters in fig.set_size_inches(w,h) and the dpi parameter in fig.savefig(fname, dpi) so that it result in 24px by 24px, it should work just fine. For example, w = h = 1 and dpi = 24 – Downpipe

I had to combine both this answer, and the answer below by Mostafa Pakparvar. Not only do you need to turn off the axes, but you need to set_visible to false to make sure the white space disappears. (wtf?) – Glassman

Just tried this using matplotlib v2.2.2 an it worked perfectly (except that imshow's syntax changed to aspect='auto' instead of 'normal'). – Yetah

This is the right approach. ax = plt.Axes(fig, [0., 0., 1., 1.]) is what makes it work. – Sharpfreeze

I had to change the last line to fig.savefig(fname, dpi=dpi). – Fortran

Answer 2

An easier solution seems to be:

import matplotlib.pyplot as plt

    plt.imshow(img)
    plt.axis('off')
    fig.savefig('out.png', bbox_inches='tight', pad_inches=0)

Necolenecro

This worked out great for me. Also, pad_inches can be changed to desired size easily. Thanks! – Candlemas

+1 worked out great for me too :) And this is actually way simpler than the accepted answer – Iraqi

I still got white margins with this. – Almswoman

Yes, for newer versions of Matplotlib the above line seemingly results in small margins. Setting the extent manually is probably the cleanest solution: fig.set_size_inches((width, height)) extent = mpl.transforms.Bbox(((0, 0), (width, height))) fig.savefig([...], bbox_inches=extent) – Necolenecro

I was able to update this for newer versions/remaining margins issue simply by adding transparent=True fig.savefig('out.png', bbox_inches='tight',transparent=True, pad_inches=0) – Helium

Still got axes and tickmarks and everything with this :( – Scintillation

This does not do anything at all. It will give you the figure with all axes and labels. – Campos

I recommend borrowing plt.axis('off') from another answer in combination with this one. – Fullblooded

Answer 3

You can find the bbox of the image inside the axis (using get_window_extent), and use the bbox_inches parameter to save only that portion of the image:

import numpy as np
import matplotlib.pyplot as plt

data=np.arange(9).reshape((3,3))
fig=plt.figure()
ax=fig.add_subplot(1,1,1)
plt.axis('off')
plt.imshow(data)

extent = ax.get_window_extent().transformed(fig.dpi_scale_trans.inverted())
plt.savefig('/tmp/test.png', bbox_inches=extent)

I learned this trick from Joe Kington here.

Perceptual

just plt.axis('off') helped. Other answers don't much help. – Commonly

This worked pretty well. However in my case there is still a small white border. Any ideas of how to remove this border? – Planchet

@user3731622, please try plt.savefig('/temp/test.png', bbox_inches='tight', transparent=True, pad_inches=0) instead of plt.savefig('/tmp/test.png', bbox_inches=extent) – Vasilikivasilis

Answer 4

I've tried several options in my case, and the best solution was this:

fig.subplots_adjust(bottom = 0)
fig.subplots_adjust(top = 1)
fig.subplots_adjust(right = 1)
fig.subplots_adjust(left = 0)

then save your figure with savefig

Subphylum

Out of all of this solutions, this is the only one that worked for me. – Cordula

Answer 5

Actually I have tried this recently and instead of all these lines, you can use

plt.imsave(image_path, image)

Works like a charm. just one line and problem solved.

imsave()

Documentation ( https://matplotlib.org/3.1.1/api/_as_gen/matplotlib.pyplot.imsave.html )

Benner

This works like a charm! It should be the best answer. – Leeth

The only problem with this approach is that, you can't use annotations and save it correctly in the image. – Furrier

@NicolásMedina please check the question again, it is about saving only content, just image, and nothing else. Its not about saving annotations or other image data. – Benner

Answer 6

I will suggest heron13 answer with a slight addition borrowed from here to remove the padding left after setting the bbox to tight mode, therefore:

axes = fig.axes()
axes.get_xaxis().set_visible(False)
axes.get_yaxis().set_visible(False)
fig.savefig('out.png', bbox_inches='tight', pad_inches=0)

Crawly

I am getting an error saying get_xaxis() and get_yaxis() don't exist. Any idea why that would happen? – Moreville

Make an axes() object, then use ax.xaxis and ax.yaxis – Pearlpearla

Got an error saying 'list' object has no attribute 'get_xaxis' – Kandace

@HaozheXie Use an Axes object, not a list of Axes... – Maverick

Answer 7

For anybody trying to do this in Jupyter Notebook

 plt.axis('off')

 spec = plt.imshow

 plt.savefig('spec',bbox_inches='tight',transparent=True, pad_inches=0)

Covenantor

Answer 8

This one work for me

plt.savefig('filename',bbox_inches='tight',transparent=True, pad_inches=0)

Recycle

Answer 9

For me, this code made similar the input image size without frame and axes. I combined snippets from matehat, unutbu, and WHZW:

fig = plt.figure()
ax = fig.add_subplot(1,1,1)
plt.axis('off')
viridis = cm.get_cmap('gist_gray', 256)
plt.imshow(data, aspect='auto', cmap=viridis)
plt.tight_layout()
plt.savefig(out_file, bbox_inches='tight', transparent=True, pad_inches=0)

Runtime environment:
  Python: 3.6.10
  Matplotlib: 3.2.1
  OS: Windows 10

Vasilikivasilis

Answer 10

I had the same problem while doing some visualization using librosa where I wanted to extract content of the plot without any other information. So this my approach. unutbu answer also helps me to make to work.

    figure = plt.figure(figsize=(500, 600), dpi=1)
    axis = plt.subplot(1, 1, 1)
    plt.axis('off')
    plt.tick_params(axis='both', left='off', top='off', right='off', bottom='off', labelleft='off', labeltop='off',
                    labelright='off', labelbottom='off')

     # your code goes here. e.g: I used librosa function to draw a image
    result = np.array(clip.feature_list['fft'].get_logamplitude()[0:2])
    librosa.display.specshow(result, sr=api.Clip.RATE, x_axis='time', y_axis='mel', cmap='RdBu_r')


    extent = axis.get_window_extent().transformed(figure.dpi_scale_trans.inverted())
    plt.savefig((clip.filename + str("_.jpg")), format='jpg', bbox_inches=extent, pad_inches=0)
    plt.close()

Guthry

This got me on the right track, but I had two problems: 1) I had to set the dpi to a number greater than 1 to avoid a font error in my jupyter notebook; and 2) there was still a small border so I have to manually change the extent Bbox to extent.get_points()*np.array([[1.1],[.9]]). – Crispa

thanks for the fulfiling the answer which may help to somebody else. – Guthry

Answer 11

While the above answers address removing margins and padding, they did not work for me in removing labels. Here's what worked, for anyone who stumbles upon this question later:

Assuming you want a 2x2 grid of subplots from four images stored in images:

matplotlib.pyplot.figure(figsize = (16,12)) # or whatever image size you require
for i in range(4):
    ax = matplotlib.pyplot.subplot(2,2,i+1)
    ax.axis('off')
    imshow(images[i])
matplotlib.pyplot.savefig(path, bbox_inches='tight')

Joannajoanne

Answer 12

I tried to get rid of the border too, using tips here but nothing really worked. Some fiddling about and I found that changing the faceolor gave me no border in jupyter labs (Any color resulted in getting rid of the white border). Hope this helps.

def show_num(data):
data = np.rot90(data.reshape((16,16)), k=3)
data = np.fliplr(data)
fig = plt.figure(frameon=False, facecolor='white')
ax = plt.Axes(fig, [0., 0., 1., 1.])
ax.set_axis_off()
fig.add_axes(ax)
ax.imshow(data)
plt.show()

Agribusiness

Answer 13

I am working on a CNN model which needs simulated images as input training data. Thus, I need axes to be removed completely. I tried all presented answers and found below combination the most effective in my case for outputting a certain size figure by pixels (in my case 512*512). As an example:

import numpy as np
import matplotlib.pyplot as plt

time = np.arange(0, 10, 0.1)
amplitude = np.sin(time)

plt.plot(amplitude, c = 'black')

plt.savefig('no_content_data.png')
plt.close()

which saves below figure in a normal way with axes and no specific size:

While below codes remove everything and output figure in 512*512 pixel size:

pixels = 512
px = 1/plt.rcParams['figure.dpi']  # pixel in inches
fig, ax = plt.subplots(frameon=False)
fig.set_size_inches(pixels*px, pixels*px)
ax = plt.Axes(fig, [0., 0., 1., 1.]) # This is needed for fitting content to 512*512 figure frame otherwise output figure will be less than desired size
ax.set_axis_off()
fig.add_axes(ax)

plt.plot(amplitude, c = 'black')

fig.subplots_adjust(bottom = 0)
fig.subplots_adjust(top = 0.00001) # This value should be very small so that marginal axes of subplot would not overlay on the main figure
fig.subplots_adjust(right = 1)
fig.subplots_adjust(left = 0)

plt.savefig('no_content_data.png')
plt.close()

which gives this:

Rutland