Generate sprite images for tensorboard

gistfile1.txt

#!/usr/bin/python

from PIL import Image
import math

test_image_list = '/home/tianwei/Data/sfm_data/test_image_list' # path to the image list
with open(test_image_list, 'r') as f:
    test_images = f.readlines()
    test_images = map(str.strip, test_images)

grid = int(math.sqrt(len(test_images))) + 1
image_height = int(8192 / grid)         # tensorboard supports sprite images up to 8192 x 8192
image_width = int(8192 / grid)

big_image = Image.new(
    mode='RGBA',
    size=(image_width * grid, image_height * grid),
    color=(0,0,0,0))  # fully transparent

for i in range(len(test_images)):
    row = i / grid
    col = i % grid
    img = Image.open(test_images[i])
    img = img.resize((image_height, image_width), Image.ANTIALIAS)
    row_loc = row * image_height
    col_loc = col * image_width
    big_image.paste(img, (col_loc, row_loc)) # NOTE: the order is reverse due to PIL saving
    print(row_loc, col_loc)

big_image.save('sprite_image.jpg', transparency=0)

This is used for making sprite images for the embedding projector.
http://projector.tensorflow.org/

Hi, I'm trying to use your code to visualize images in tensorflow projector. PIL does not allow to save a .jpg image in RGBA format (the transparency channel is not supported). Any idea on how to approach this? The sprite looks nice, but I cannot get it to work in the projector.

Specifically, I get the "Number of tensors (196) do not match the number of lines in metadata (202)." error when trying to load the sprite as metadata. I am trying to plot 196 images.

Have you found a solution to this problem? I'm also very positive that I've created the sprite the right way, but I just can't find a way of uploading it. If I upload it instead of the .tsv I get the number of tensors/lines mismatch error.

This worked for me.
changed:

1. RGBA --> RGB
2. color=(0,0,0,0) --> color=(0,0,0)
3. row = i / grid --> row = i // grid
4. big_image.save('sprite_image.jpg', transparency=0) --> big_image.save('sprite_image.jpg')

big_image = Image.new( mode='RGB',
                       size=(image_width * grid, image_height * grid),
                       color=(0,0,0) )  # reduced it to RGB

for i in range(len(test_images)):

    row     = i // grid  # added integer divide
    col     = i % grid
    img     = Image.open(test_images[i])
    img     = img.resize((image_height, image_width), Image.ANTIALIAS)
    row_loc = row * image_height
    col_loc = col * image_width

    big_image.paste(img, (col_loc, row_loc)) # NOTE: the order is reverse due to PIL saving
    print(row_loc, col_loc)

big_image.save('sprite_image.jpg') # removed transparency