Myujth

I need to submit PNG-images to a blackbox server with a http request. I use python3 to generate images in a numpy 64x64x3 array. What I currently do is:

1. Generate image


2. Save image with scipy.misc.toimage to disk


3. Open saved image file from disk


4. Use requests module to send http requests with the image opened image file object in it

This works perfectly fine, but I want to get rid of step 2 and 3, so I do not need to save my object to disk first and then load it again. Instead I would like to convert my numpy array in a file-object that is compatible with the http server and send it directly. (Like one you get from open() )

I know it is easy to convert from numpy array to PNG-image with PIL for example, but I only find how to do that combined with saving to disk in one function.

Thank you very much for any help!

This is my code so far:

import numpy as np
import requests
from scipy.misc import toimage

arr = generate64x64x3ImageWithNumpy()
toimage(arr, cmin=0.0, cmax=255.0).save('tmp.png')
d = 'key':API_KEY
f= 'image': open('tmp.png', 'rb')
result = requests.post(SERVER_URL, files=f, data=d)

I want this:

arr = generate64x64x3ImageWithNumpy()

not_on_disk = numpyArrayToPNGImageWithoutSavingOnDisk(arr)

d = 'key':API_KEY
f = 'image': not_on_disk
result = requests.post(SERVER_URL, files=f, data=d)

You can use an in memory iostream with savefig (https://docs.python.org/3/library/io.html#io.BytesIO)

import io
tmpFile = io.BytesIO()
savefig(tmpFile, format='png')

To verify that this worked tmpFile can be compared with the actual file as saved to disk.

# Get contents of tmpFile
tmpFile.seek(0)
not_on_disk = tmpFile.read(-1)

# Save to and load from disk
fname = 'tmp.png'
savefig(fname)
on_disk = open(fname, 'rb').read(-1)

>>>not_on_disk == on_disk
True

Edit You're looking at using scipy and pil rather than matplotlib but the answer should work the same, including the format keyword for saving.