feat: update web Interface
This commit is contained in:
parent
d5d633b6c7
commit
f9844dba10
1111 changed files with 171093 additions and 0 deletions
48
WebInterface/Front-end/generate-experiment-files/blur_vis.py
Normal file
48
WebInterface/Front-end/generate-experiment-files/blur_vis.py
Normal file
|
|
@ -0,0 +1,48 @@
|
|||
# -*- coding: utf-8 -*-
|
||||
"""
|
||||
Created on Thu Jul 8 17:53:59 2021
|
||||
|
||||
@author: jiao1
|
||||
"""
|
||||
|
||||
from PIL import Image, ImageFilter
|
||||
from pathlib import Path
|
||||
|
||||
current_path = Path('./')
|
||||
|
||||
for i in range(1,11):
|
||||
p = current_path.joinpath(str(i))
|
||||
out_path = current_path.joinpath(str(i)+'_blur')
|
||||
if not out_path.exists():
|
||||
out_path.mkdir()
|
||||
for file_name in p.rglob('*.png'):
|
||||
name = str(file_name).split("/")[-1]
|
||||
file_name = str(file_name)
|
||||
image = Image.open(file_name)
|
||||
image = image.convert(mode="RGB")
|
||||
print(image.size)
|
||||
|
||||
if image.size[0]>2000 or image.size[1]>2000:
|
||||
blurred = image.filter(ImageFilter.GaussianBlur(radius = 24))
|
||||
elif image.size[0]>1000 or image.size[1]>1000:
|
||||
blurred = image.filter(ImageFilter.GaussianBlur(radius = 12))
|
||||
else:
|
||||
blurred = image.filter(ImageFilter.GaussianBlur(radius = 5))
|
||||
out = out_path.joinpath(name)
|
||||
blurred.save(str(out))
|
||||
for file_name in p.rglob('*.jpg'):
|
||||
name = str(file_name).split("\\")[-1]
|
||||
file_name = str(file_name)
|
||||
image = Image.open(file_name)
|
||||
image = image.convert(mode="RGB")
|
||||
print(image.size)
|
||||
|
||||
if image.size[0]>2000 or image.size[1]>2000:
|
||||
blurred = image.filter(ImageFilter.GaussianBlur(radius = 24))
|
||||
elif image.size[0]>1000 or image.size[1]>1000:
|
||||
blurred = image.filter(ImageFilter.GaussianBlur(radius = 12))
|
||||
else:
|
||||
blurred = image.filter(ImageFilter.GaussianBlur(radius = 5))
|
||||
out = out_path.joinpath(name)
|
||||
blurred.save(str(out))
|
||||
|
||||
|
|
@ -0,0 +1,40 @@
|
|||
from PIL import Image
|
||||
import os
|
||||
import sys
|
||||
|
||||
BASE = "../assets/task_data/"
|
||||
|
||||
dirs = ["all_images", "sentinel_CC", "natural", "tutorial_CC", "real_CC", "sentinel_images", "tutorial_sentinels", "real_images", "tutorial_images"]
|
||||
|
||||
def get_bad_ims(base, dirs):
|
||||
n = 0
|
||||
bad = []
|
||||
dirs_to_check = [os.path.join(base, d) for d in dirs]
|
||||
for d in dirs_to_check:
|
||||
for path, subdirs, files in os.walk(d):
|
||||
for name in files:
|
||||
im_path = os.path.join(path, name)
|
||||
n += 1
|
||||
if not _is_valid_img(im_path):
|
||||
bad.append(name)
|
||||
print("Checked %d imgs" % n)
|
||||
return bad
|
||||
|
||||
|
||||
def _is_valid_img(im_path):
|
||||
try:
|
||||
Image.open(im_path)
|
||||
return True
|
||||
except:
|
||||
return False
|
||||
|
||||
if __name__ == "__main__":
|
||||
#print(_is_valid_img("valid.jpg"))
|
||||
base = None
|
||||
if len(sys.argv) == 1:
|
||||
print("Using % as base path" % BASE)
|
||||
base = BASE
|
||||
else:
|
||||
base = sys.argv[1]
|
||||
print(base)
|
||||
print(get_bad_ims(base, dirs))
|
||||
|
|
@ -0,0 +1,50 @@
|
|||
import generate_codecharts
|
||||
import os
|
||||
import json
|
||||
|
||||
|
||||
def create_codecharts(real_CC_dir,ncodecharts,image_width,image_height):
|
||||
|
||||
# new directory (will get populated in this file)
|
||||
if not os.path.exists(real_CC_dir):
|
||||
os.makedirs(real_CC_dir)
|
||||
# note: no buckets in this directory, all buckets will sample from a single source
|
||||
|
||||
data = {}
|
||||
data_with_coords = {}
|
||||
print('Generating %d codecharts...'%(ncodecharts))
|
||||
for img_num in range(ncodecharts):
|
||||
if img_num%100==0:
|
||||
print('%d/%d'%(img_num,ncodecharts))
|
||||
filename = os.path.join(real_CC_dir,'real_CC_%d.jpg'%(img_num))
|
||||
valid_codes, coordinates = generate_codecharts.create_codechart(filename,image_width,image_height)
|
||||
data[filename] = {'valid_codes':valid_codes}
|
||||
data_with_coords[filename] = {'valid_codes':valid_codes, 'coordinates':coordinates}
|
||||
|
||||
with open(os.path.join(real_CC_dir,'CC_codes.json'), 'w') as outfile:
|
||||
json.dump(data, outfile)
|
||||
print('Writing out %s'%(os.path.join(real_CC_dir,'CC_codes.json')))
|
||||
|
||||
with open(os.path.join(real_CC_dir,'CC_codes_full.json'), 'w') as outfile:
|
||||
json.dump(data_with_coords, outfile)
|
||||
print('Writing out %s'%(os.path.join(real_CC_dir,'CC_codes_full.json')))
|
||||
|
||||
print('Done!')
|
||||
|
||||
if __name__ == "__main__":
|
||||
|
||||
rootdir = './task_data'
|
||||
|
||||
# use these settings to figure out how many codecharts to make
|
||||
# so that each codechart will be sampled once
|
||||
num_subject_files = 200 #100 # default: 100 subject files/ bucket
|
||||
num_images_per_sf = 35 #20 # default: 20 images/ subject file
|
||||
ncodecharts = 20 #num_subject_files*num_images_per_sf
|
||||
|
||||
image_width = 1920 # in pixel
|
||||
image_height = 1080 # in pixel
|
||||
|
||||
real_CC_dir = os.path.join(rootdir,'real_CC')
|
||||
|
||||
create_codecharts(real_CC_dir,ncodecharts,image_width,image_height)
|
||||
|
||||
|
|
@ -0,0 +1,89 @@
|
|||
import os
|
||||
import glob
|
||||
from PIL import Image
|
||||
from collections import Counter
|
||||
import matplotlib.pyplot as plt
|
||||
|
||||
# resize all images to these max dimensions (to be consistent across experiments with different image sizes)
|
||||
MAX_H = 1340 #1920 1920 resized to 1000, max height can be 700 so 700x(1920/1000)
|
||||
MAX_W = 1920
|
||||
to_resize = True
|
||||
todebug = False
|
||||
|
||||
def get_max_dims(allfiles):
|
||||
|
||||
widths = []
|
||||
heights = []
|
||||
for file in allfiles:
|
||||
im = Image.open(file)
|
||||
width, height = im.size
|
||||
widths.append(width)
|
||||
heights.append(height)
|
||||
|
||||
if todebug:
|
||||
print("Image widths:",Counter(widths).keys())
|
||||
print("Image heights:",Counter(heights).keys())
|
||||
|
||||
maxwidth = max(list(Counter(widths).keys()))
|
||||
maxheight = max(list(Counter(heights).keys()))
|
||||
|
||||
if to_resize:
|
||||
ratio = min(MAX_W/maxwidth, MAX_H/maxheight)
|
||||
maxwidth = int(maxwidth*ratio)
|
||||
maxheight = int(maxheight*ratio)
|
||||
|
||||
return maxwidth,maxheight
|
||||
|
||||
def save_padded_images(real_image_dir,allfiles,toplot=False,maxwidth=None,maxheight=None):
|
||||
|
||||
if maxwidth==None or maxheight==None:
|
||||
maxwidth,maxheight = get_max_dims(allfiles)
|
||||
|
||||
print('Padding %d image files to dimensions: [%d,%d]...'%(len(allfiles),maxwidth,maxheight))
|
||||
|
||||
for file in allfiles:
|
||||
#print(file)
|
||||
im = Image.open(file)
|
||||
|
||||
if to_resize:
|
||||
#resize image to fixed dimensions
|
||||
width, height = im.size
|
||||
ratio = min(maxwidth/width, maxheight/height)
|
||||
newwidth = int(width*ratio)
|
||||
newheight = int(height*ratio)
|
||||
if todebug: print('resizing %dx%d to %dx%d (ratio=%2.2f)'%(width,height,newwidth,newheight,ratio))
|
||||
im = im.resize((newwidth,newheight), Image.ANTIALIAS)
|
||||
|
||||
width, height = im.size
|
||||
padded_im = Image.new('RGB',
|
||||
(maxwidth, maxheight),
|
||||
(126, 126, 126))
|
||||
offset = ((maxwidth - width) // 2, (maxheight - height) // 2)
|
||||
padded_im.paste(im, offset)
|
||||
padded_im.save(os.path.join(real_image_dir,os.path.basename(file)))
|
||||
|
||||
if toplot:
|
||||
plt.figure(figsize=(10,10))
|
||||
plt.imshow(padded_im)
|
||||
plt.axis('off');
|
||||
plt.show();
|
||||
|
||||
print('Done!')
|
||||
return maxwidth,maxheight
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
|
||||
sourcedir = '../../importance_dataset/predimportance/analysis/CVs_all' # take images from here
|
||||
rootdir = './task_data'
|
||||
|
||||
real_image_dir = os.path.join(rootdir,'real_images')
|
||||
if not os.path.exists(real_image_dir):
|
||||
print('Creating directory %s'%(real_image_dir))
|
||||
os.makedirs(real_image_dir)
|
||||
|
||||
allfiles = glob.glob(os.path.join(sourcedir,'*.png'))
|
||||
|
||||
save_padded_images(real_image_dir,allfiles,toplot=False)
|
||||
|
||||
|
||||
|
|
@ -0,0 +1,39 @@
|
|||
import os
|
||||
import random
|
||||
from shutil import copyfile
|
||||
import glob
|
||||
|
||||
def distribute_images(from_dir,real_image_dir,num_buckets,start_bucket_at):
|
||||
# distribute images across buckets
|
||||
|
||||
img_files = []
|
||||
for ext in ('*.jpeg', '*.png', '*.jpg'):
|
||||
img_files.extend(glob.glob(os.path.join(from_dir, ext)))
|
||||
|
||||
random.shuffle(img_files) # shuffle all images randomly at start
|
||||
images_per_bucket = int(len(img_files)/float(num_buckets))
|
||||
|
||||
print('Distributing images across %d buckets'%(num_buckets))
|
||||
for b in range(num_buckets):
|
||||
bucket_dir = os.path.join(real_image_dir,'bucket%d'%(start_bucket_at+b))
|
||||
if not os.path.exists(bucket_dir):
|
||||
os.makedirs(bucket_dir)
|
||||
print('Populating %s/bucket%d with %d images'%(real_image_dir,start_bucket_at+b,images_per_bucket))
|
||||
for i in range(b*images_per_bucket,b*images_per_bucket+images_per_bucket):
|
||||
destfile = os.path.basename(img_files[i])
|
||||
copyfile(img_files[i], os.path.join(bucket_dir,destfile))
|
||||
|
||||
if __name__ == "__main__":
|
||||
|
||||
num_buckets = 1 # num buckets to split images into
|
||||
start_bucket_at = 0 # where to start the naming of the buckets (in case other buckets already exist)
|
||||
|
||||
rootdir = './task_data'
|
||||
|
||||
real_image_dir = os.path.join(rootdir,'real_images')
|
||||
if not os.path.exists(real_image_dir):
|
||||
os.makedirs(real_image_dir)
|
||||
|
||||
from_dir = os.path.join(rootdir,'all_images')
|
||||
|
||||
distribute_images(from_dir,real_image_dir,num_buckets,start_bucket_at)
|
||||
|
|
@ -0,0 +1,39 @@
|
|||
from PIL import Image, ImageDraw, ImageColor, ImageFont
|
||||
import matplotlib.pyplot as plt
|
||||
import numpy as np
|
||||
import json
|
||||
import os
|
||||
|
||||
# FIXED parameters
|
||||
text_color = ImageColor.getrgb("white")
|
||||
font_type = "arial.ttf"
|
||||
px_pt_ratio = 20/29 # according to our image dimensions, 29 point = 20 px
|
||||
|
||||
def pixel_to_point(num):
|
||||
return int(num*(1/px_pt_ratio))
|
||||
|
||||
def save_fixation_cross(rootdir,image_width,image_height):
|
||||
|
||||
font_size = int(image_height*0.0278)
|
||||
print('using font size: %d'%(font_size))
|
||||
|
||||
# Generate and save fixation cross image
|
||||
img = Image.new('RGB', (image_width, image_height), (126, 126, 126))
|
||||
d = ImageDraw.Draw(img)
|
||||
try:
|
||||
font = ImageFont.truetype(font_type, pixel_to_point(font_size)) # takes in point value
|
||||
except OSError:
|
||||
print("WARNING: using a different font because oculd not find %s on your computer" % font_type)
|
||||
font = ImageFont.truetype("/usr/share/fonts/truetype/freefont/FreeSans.ttf", pixel_to_point(font_size))
|
||||
d.text((image_width/2.0 - font_size, image_height/2.0 - font_size), '+', text_color, font)
|
||||
filename = 'fixation-cross.jpg'
|
||||
img.save(os.path.join(rootdir,filename))
|
||||
print('Saved fixation cross image as %s'%(os.path.join(rootdir,filename)))
|
||||
|
||||
if __name__ == "__main__":
|
||||
|
||||
rootdir = './task_data'
|
||||
image_width = 1920 # in pixel
|
||||
image_height = 1080 # in pixel
|
||||
|
||||
|
||||
|
|
@ -0,0 +1,166 @@
|
|||
from PIL import Image, ImageDraw, ImageColor, ImageFont
|
||||
import matplotlib.pyplot as plt
|
||||
import numpy as np
|
||||
import string
|
||||
import random
|
||||
import json
|
||||
import os
|
||||
import math
|
||||
|
||||
# DEFINE PARAMATERS
|
||||
forbidden_letters = set(["I", "O"]) # letters to not use in code charts because can be confused with digits
|
||||
px_pt_ratio = 20/29 # according to our image dimensions, 29 point = 20 px
|
||||
text_color = ImageColor.getrgb("gray")
|
||||
font_type = "arial.ttf"
|
||||
tojitter = True # add jitter from a regular grid
|
||||
ebuf = 5 # buffer number of pixels to leave from the edges of the image so codecharts are not tangent to image edges
|
||||
go_to_image_edges = False # if want to make sure to sample triplets to the very edge of the image (downside: triplets may be more crowded)
|
||||
|
||||
def point_to_pixel(num):
|
||||
return int(num*px_pt_ratio)
|
||||
|
||||
def pixel_to_point(num):
|
||||
return int(num*(1/px_pt_ratio))
|
||||
|
||||
def generate_rand_letter():
|
||||
letter = random.choice(string.ascii_uppercase)
|
||||
while letter in forbidden_letters:
|
||||
letter = random.choice(string.ascii_uppercase)
|
||||
return letter
|
||||
|
||||
def generate_rand_triplet():
|
||||
code = ""
|
||||
code += generate_rand_letter()
|
||||
# the following code prevents the two digits from being identical or equal to 0
|
||||
for i in range(2):
|
||||
if i == 0:
|
||||
forbidden_num = 0
|
||||
else:
|
||||
forbidden_num = int(code[i])
|
||||
r = list(range(1, forbidden_num)) + list(range(forbidden_num+1, 10))
|
||||
code += str(random.choice(r))
|
||||
return code
|
||||
|
||||
def create_codechart(filename,image_width,image_height):
|
||||
|
||||
font_size = int(image_height*0.0185)
|
||||
|
||||
# all these parameters depend on font size
|
||||
max_triplet_width = font_size*3 # in pixel - max triplet width; used 'W88' as widest triplet code (width~60, height=20)
|
||||
max_triplet_height = font_size # the tallest a triplet can be
|
||||
d_v = 4*max_triplet_height # vertical distance to maintain b/w triplets in the grid
|
||||
d_h = 2*max_triplet_width # horizontal distance to maintain b/w triplets (from start of one triplet to start of another)
|
||||
|
||||
# make sure that not too much empty space is left over by spacing out triplets
|
||||
N_h = int(math.floor((image_width-max_triplet_width-2*ebuf) / float(d_h))) # number of triplets that will be tiled horizontally
|
||||
d_h = int(math.floor((image_width-max_triplet_width-2*ebuf) / float(N_h))) # recompute the horizontal dist between triplets to eliminate extra space
|
||||
N_v = int(math.floor((image_height-max_triplet_height-2*ebuf) / float(d_v)))
|
||||
d_v = int(math.floor((image_height-max_triplet_height-2*ebuf) / float(N_v)))
|
||||
# -------------
|
||||
|
||||
post_jitter_buffer = 6 # small buffer to cover edge case of triplets immediately adjacent to one another (for legibility)
|
||||
j_v = int(0.25*(d_v) - post_jitter_buffer/2) # max vertical jitter for one side of a triplet
|
||||
j_h = int(0.25*(d_h) - post_jitter_buffer) # max horizontal jitter for on side of a triplet
|
||||
|
||||
# set up image canvas and font size/style
|
||||
img = Image.new('RGB', (image_width, image_height))
|
||||
d = ImageDraw.Draw(img)
|
||||
try:
|
||||
font = ImageFont.truetype(font_type, pixel_to_point(font_size)) # takes in point value
|
||||
except OSError:
|
||||
print("WARNING: using a different font bc could not find %s on your computer" % font_type)
|
||||
font = ImageFont.truetype("/usr/share/fonts/truetype/freefont/FreeSans.ttf", pixel_to_point(font_size))
|
||||
|
||||
valid_codes = set()
|
||||
coordinates = {}
|
||||
|
||||
# initialize starting locations for triplets on the image grid
|
||||
x_init = ebuf
|
||||
y_init = ebuf
|
||||
|
||||
# -------- improvement made after 01/2020 (after TurkEyes paper) --------
|
||||
# original problem was grid-like artifacts in the collected data because the grid spacing between consecutive triplets
|
||||
# was always similar (despite a small bit of jitter added when using the triplet)
|
||||
xoffset = random.choice(list(range(int(d_h/2.0))))
|
||||
yoffset = random.choice(list(range(int(d_v/2.0))))
|
||||
x_init = x_init+xoffset
|
||||
y_init = y_init+yoffset
|
||||
# -----------------------------------------------------------------------
|
||||
|
||||
x = x_init
|
||||
while x < image_width-max_triplet_width-ebuf:
|
||||
|
||||
y = y_init
|
||||
while y < image_height-max_triplet_height-ebuf:
|
||||
|
||||
triplet_code = generate_rand_triplet()
|
||||
|
||||
# check for if triplet has already been used in image since all codes should be unique
|
||||
while triplet_code in valid_codes:
|
||||
triplet_code = generate_rand_triplet()
|
||||
valid_codes.add(triplet_code)
|
||||
|
||||
if tojitter:
|
||||
# implement jitter to x and y coordinates (note: can turn either of them off)
|
||||
min_x = max(ebuf,x-j_h)
|
||||
max_x = min(x+j_h+1,image_width-max_triplet_width-ebuf)
|
||||
min_y = max(ebuf,y-j_v)
|
||||
max_y = min(y+j_v+1,image_height-max_triplet_height-ebuf-2) # a little bit of extra buffer in vertical dimension
|
||||
x_range = list(range(min_x, max_x))
|
||||
y_range = list(range(min_y, max_y))
|
||||
j_x = random.choice(x_range)
|
||||
j_y = random.choice(y_range)
|
||||
else:
|
||||
j_x = x
|
||||
j_y = y
|
||||
|
||||
# writes triplet to image
|
||||
d.text((j_x, j_y), triplet_code, text_color, font)
|
||||
coordinates[triplet_code] = (j_x, j_y)
|
||||
|
||||
y_prev = y
|
||||
y = y+d_v # regularly sample the image vertically
|
||||
|
||||
# triplets are not guaranteed to go to edge of image, and gap could be large
|
||||
# see if can still squeeze in a triplet without overlapping previous ones (could still be quite close)
|
||||
if go_to_image_edges and y >= image_height-max_triplet_height-ebuf:
|
||||
y = y_prev + max_triplet_height+j_v+1 + post_jitter_buffer*2
|
||||
|
||||
x_prev = x
|
||||
x = x+d_h # regularly sample the image horizontally
|
||||
|
||||
if go_to_image_edges and x >= image_width-max_triplet_width-ebuf:
|
||||
x = x_prev + max_triplet_width+j_h+1 + post_jitter_buffer*2
|
||||
|
||||
|
||||
img.save(filename)
|
||||
return (list(valid_codes), coordinates)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
# create some code charts to test this code
|
||||
rootdir = './task_data'
|
||||
num_codecharts = 3 # generate this many codecharts
|
||||
|
||||
#image_width = 1920 # in pixel
|
||||
#image_height = 1080 # in pixel
|
||||
|
||||
image_height = 1340 #1344
|
||||
image_width = int(1036*image_height/float(1344))
|
||||
|
||||
# set up directories
|
||||
if not os.path.exists(rootdir):
|
||||
os.makedirs(rootdir)
|
||||
test_dir = os.path.join(rootdir,'TEST')
|
||||
if not os.path.exists(test_dir):
|
||||
os.makedirs(test_dir)
|
||||
|
||||
data = {}
|
||||
for i in range(num_codecharts):
|
||||
filename = os.path.join(test_dir,'CC_%d.jpg'%(i))
|
||||
valid_codes, coordinates = create_codechart(filename,image_width,image_height)
|
||||
data[filename] = (valid_codes, coordinates)
|
||||
|
||||
with open(os.path.join(test_dir,'data.json'), 'w') as outfile:
|
||||
json.dump(data, outfile)
|
||||
|
||||
|
|
@ -0,0 +1,151 @@
|
|||
from PIL import Image, ImageDraw, ImageColor, ImageFont
|
||||
import matplotlib.pyplot as plt
|
||||
import numpy as np
|
||||
import string
|
||||
import random
|
||||
import json
|
||||
import generate_codecharts as gc
|
||||
import os
|
||||
|
||||
# FIXED parameters
|
||||
text_color = ImageColor.getrgb("white")
|
||||
font_type = "arial.ttf"
|
||||
px_pt_ratio = 20/29 # according to our image dimensions, 29 point = 20 px
|
||||
valid_target_types=["red_dot", "fix_cross", "img"]
|
||||
|
||||
def make_sentinel(codechart_filename,sentinel_filename,image_width,image_height,border_padding,target_type="red_dot", target_im_dir=""):
|
||||
# border_padding used to guarantee that chosen sentinel location is not too close to border to be hard to spot
|
||||
|
||||
font_size = int(image_height*0.0278)
|
||||
correct_codes = []
|
||||
|
||||
if target_type not in valid_target_types:
|
||||
raise RuntimeError("target_type must be one of %s" % valid_target_types.__str__())
|
||||
valid_codes, coordinates = gc.create_codechart(codechart_filename,image_width,image_height)
|
||||
# pick random code
|
||||
r = list(range(0, len(valid_codes)))
|
||||
index = random.choice(r)
|
||||
triplet = valid_codes[index]
|
||||
triplet_coordinate = coordinates[triplet]
|
||||
# to make sure that the cross is visible
|
||||
while (triplet_coordinate[0] <= border_padding or triplet_coordinate[0] >= image_width-border_padding) \
|
||||
or (triplet_coordinate[1] <= border_padding or triplet_coordinate[1] >=image_height-border_padding):
|
||||
index = random.choice(r)
|
||||
triplet = valid_codes[index]
|
||||
triplet_coordinate = coordinates[triplet]
|
||||
# check bg color
|
||||
if target_type == "fix_cross":
|
||||
bg_color = 126
|
||||
else:
|
||||
bg_color = 255
|
||||
# create and save cross sentinel image
|
||||
img = Image.new('RGB', (image_width, image_height), (bg_color, bg_color, bg_color))
|
||||
d = ImageDraw.Draw(img)
|
||||
try:
|
||||
font = ImageFont.truetype(font_type, gc.pixel_to_point(font_size)) # takes in point value
|
||||
except OSError:
|
||||
print("WARNING: using different font bc could not find %s" % font_type)
|
||||
font = ImageFont.truetype("/usr/share/fonts/truetype/freefont/FreeSans.ttf", gc.pixel_to_point(font_size))
|
||||
|
||||
if target_type == "fix_cross":
|
||||
plot_coord = (triplet_coordinate[0]+font_size, triplet_coordinate[1]) # offset cross location to the center of the triplet
|
||||
d.text(plot_coord, '+', text_color, font)
|
||||
elif target_type == "red_dot":
|
||||
d.ellipse((triplet_coordinate[0], triplet_coordinate[1], triplet_coordinate[0]+font_size*2, triplet_coordinate[1]+font_size*2), fill = 'red', outline ='red')
|
||||
elif target_type == "img":
|
||||
if not target_im_dir:
|
||||
raise RuntimeError("No im dir provided for sentinel targets")
|
||||
# Get a list of images in the target im dir
|
||||
images = os.listdir(target_im_dir)
|
||||
target = Image.open(os.path.join(target_im_dir, random.choice(images)))
|
||||
# resize the target
|
||||
width = 200
|
||||
height = int(target.height*width/target.width)
|
||||
target = target.resize((width, height))
|
||||
plot_coord = (triplet_coordinate[0]-int(width/2), triplet_coordinate[1]-int(height/2))
|
||||
img.paste(target, plot_coord)
|
||||
|
||||
# correct_codes lie within the sentinel width
|
||||
for ii in range(len(valid_codes)):
|
||||
dist = np.linalg.norm(np.array(coordinates[valid_codes[ii]])-np.array(triplet_coordinate))
|
||||
if dist <= width/2.0:
|
||||
correct_codes.append(valid_codes[ii]);
|
||||
|
||||
pass
|
||||
else:
|
||||
raise RuntimeError("target_type %s does not exist" % target_type)
|
||||
img.save(sentinel_filename)
|
||||
D = {'correct_code':triplet, 'coordinate':triplet_coordinate, 'correct_codes':correct_codes}
|
||||
D_full = {'correct_code':triplet, 'coordinate':triplet_coordinate, \
|
||||
'valid_codes':valid_codes, 'coordinates':coordinates, 'codechart_file':codechart_filename,\
|
||||
'correct_codes':correct_codes}
|
||||
return D,D_full
|
||||
|
||||
def generate_sentinels(sentinel_image_dir,sentinel_CC_dir,num_buckets,start_bucket_at,sentinel_images_per_bucket,\
|
||||
image_width,image_height,border_padding,target_type,target_im_dir=""):
|
||||
|
||||
# Set up directories
|
||||
if not os.path.exists(sentinel_image_dir):
|
||||
os.makedirs(sentinel_image_dir)
|
||||
if not os.path.exists(sentinel_CC_dir):
|
||||
os.makedirs(sentinel_CC_dir)
|
||||
|
||||
# Start generating sentinels
|
||||
img_num_offset = (start_bucket_at-1)*sentinel_images_per_bucket # start at a new index id
|
||||
for b in range(num_buckets):
|
||||
|
||||
image_bucket_dir = os.path.join(sentinel_image_dir,'bucket%d'%(start_bucket_at+b))
|
||||
if not os.path.exists(image_bucket_dir):
|
||||
os.makedirs(image_bucket_dir)
|
||||
|
||||
CC_bucket_dir = os.path.join(sentinel_CC_dir,'bucket%d'%(start_bucket_at+b))
|
||||
if not os.path.exists(CC_bucket_dir):
|
||||
os.makedirs(CC_bucket_dir)
|
||||
|
||||
data = {} # save to a json the filename, the coordinate of the + cross, and the triplet at that coordinate
|
||||
data_with_coords = {} # also save a list of other valid triplets and coordinates (for analysis)
|
||||
|
||||
print('Populating %s with %d sentinel images'%(image_bucket_dir,sentinel_images_per_bucket))
|
||||
print('Populating %s with %d corresponding codecharts'%(CC_bucket_dir,sentinel_images_per_bucket))
|
||||
for i in range(sentinel_images_per_bucket):
|
||||
img_num = img_num_offset + b*sentinel_images_per_bucket + i + 1
|
||||
# generate random code chart
|
||||
codechart_filename = os.path.join(CC_bucket_dir,'sentinel_CC_%d.jpg'%(img_num))
|
||||
sentinel_filename = os.path.join(image_bucket_dir,'sentinel_image_%d.jpg'%(img_num))
|
||||
|
||||
D,D_full = make_sentinel(codechart_filename,sentinel_filename,image_width,image_height,border_padding,target_type, target_im_dir)
|
||||
|
||||
data[sentinel_filename] = D
|
||||
data_with_coords[sentinel_filename] = D_full
|
||||
|
||||
with open(os.path.join(image_bucket_dir,'sentinel_codes.json'), 'w') as outfile:
|
||||
json.dump(data, outfile)
|
||||
print('Writing out %s'%(os.path.join(image_bucket_dir,'sentinel_codes.json')))
|
||||
|
||||
with open(os.path.join(image_bucket_dir,'sentinel_codes_full.json'), 'w') as outfile:
|
||||
json.dump(data_with_coords, outfile)
|
||||
print('Writing out %s'%(os.path.join(image_bucket_dir,'sentinel_codes_full.json')))
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
|
||||
# Set these parameters
|
||||
sentinel_images_per_bucket = 500
|
||||
num_buckets = 1
|
||||
start_bucket_at = 0 # to avoid overwriting the existing buckets
|
||||
|
||||
image_width = 1920
|
||||
image_height = 1080
|
||||
border_padding = 100 # don't put fixation cross in this region of the image
|
||||
rootdir = './task_data'
|
||||
|
||||
target_type = "img"
|
||||
target_im_dir = "sentinel_target_images"
|
||||
|
||||
sentinel_image_dir = os.path.join(rootdir,'sentinel_images')
|
||||
sentinel_CC_dir = os.path.join(rootdir,'sentinel_CC')
|
||||
|
||||
generate_sentinels(sentinel_image_dir,sentinel_CC_dir,num_buckets,start_bucket_at,sentinel_images_per_bucket,\
|
||||
image_width,image_height,border_padding,target_type, target_im_dir)
|
||||
|
||||
|
||||
|
|
@ -0,0 +1,123 @@
|
|||
from cv2 import cv2
|
||||
import numpy as np
|
||||
import re,os
|
||||
from glob import glob
|
||||
from PIL import Image
|
||||
import matplotlib.pyplot as plt
|
||||
from shutil import rmtree
|
||||
import pandas as pd
|
||||
from natsort import natsorted
|
||||
import imageio
|
||||
import moviepy.editor as mp
|
||||
|
||||
bp = "E:/1Study/Hiwi/Massvis dataset" # change base path to the directory where you have downloaded the salicon data
|
||||
curdir = 'E:/1Study/Hiwi/massvis/eyetracking/csv_files/fixationsByVis'
|
||||
eledir = 'E:/1Study/Hiwi/massvis/dataset/csv_files/targets393_elementLabels/elementLabels/'
|
||||
origimdir = os.path.join(bp, 'targets')
|
||||
|
||||
def highlight_function(x,y,r,imagePath,outPath,imgName):
|
||||
# load input image
|
||||
img= cv2.imread(imagePath, cv2.IMREAD_UNCHANGED)
|
||||
|
||||
# blur the image_origin to imgBlur
|
||||
imgBlur=cv2.blur(img,(15,15))
|
||||
|
||||
# reduce brightness of imgBlur by 40%
|
||||
w=imgBlur.shape[1]
|
||||
h=imgBlur.shape[0]
|
||||
|
||||
imgBrightness=0.6
|
||||
|
||||
for xi in range(0,w):
|
||||
for xj in range(0,h):
|
||||
imgBlur[xj,xi,0]=int(imgBlur[xj,xi,0]*imgBrightness)
|
||||
imgBlur[xj,xi,1]=int(imgBlur[xj,xi,1]*imgBrightness)
|
||||
imgBlur[xj,xi,2]=int(imgBlur[xj,xi,2]*imgBrightness)
|
||||
|
||||
|
||||
# get size of image
|
||||
height, width = img.shape[:2]
|
||||
height = int(height)
|
||||
width = int(width)
|
||||
|
||||
# generate in-circle-display template
|
||||
circleIn = np.zeros((height, width, 1), np.uint8)
|
||||
circleIn = cv2.circle(circleIn, (x, y), r, (1), -1)
|
||||
|
||||
# generate out-circle-display template
|
||||
circleOut = circleIn.copy()
|
||||
circleOut[circleOut == 0] = 2
|
||||
circleOut[circleOut == 1] = 0
|
||||
circleOut[circleOut == 2] = 1
|
||||
|
||||
# generate imgIn in which only the content in the spotlight circle remains
|
||||
# generate a blank img with the same size of input image
|
||||
imgIn = np.zeros((height, width, 4), np.uint8)
|
||||
# copy first 3 channel
|
||||
imgIn[:, :, 0] = np.multiply(img[:, :, 0], circleIn[:, :, 0])
|
||||
imgIn[:, :, 1] = np.multiply(img[:, :, 1], circleIn[:, :, 0])
|
||||
imgIn[:, :, 2] = np.multiply(img[:, :, 2], circleIn[:, :, 0])
|
||||
# set non-transparent part of α channel
|
||||
circleIn[circleIn == 1] = 255
|
||||
imgIn[:, :, 3] = circleIn[:, :, 0]
|
||||
|
||||
# generate imgOut in which only the content outside the spotlight circle remains and be blurred
|
||||
imgOut = np.zeros((height, width, 4), np.uint8)
|
||||
imgOut[:, :, 0] = np.multiply(imgBlur[:, :, 0], circleOut[:, :, 0])
|
||||
imgOut[:, :, 1] = np.multiply(imgBlur[:, :, 1], circleOut[:, :, 0])
|
||||
imgOut[:, :, 2] = np.multiply(imgBlur[:, :, 2], circleOut[:, :, 0])
|
||||
circleOut[circleOut == 1] = 255
|
||||
imgOut[:, :, 3] = circleOut[:, :, 0]
|
||||
|
||||
# generate output image by adding imgIn and imgOut
|
||||
imgHighlight = cv2.add(imgIn,imgOut)
|
||||
cv2.imwrite(os.path.join(outPath , imgName+'.png'),imgHighlight)
|
||||
|
||||
if(cv2.waitKey(0)==27):
|
||||
cv2.destroyAllWindows()
|
||||
|
||||
return imgHighlight
|
||||
|
||||
|
||||
for curfile in os.listdir(curdir):
|
||||
os.makedirs(os.path.join(bp, 'highlight'), exist_ok=True)
|
||||
outpath = os.path.join(bp, 'highlight', curfile)
|
||||
os.makedirs(outpath, exist_ok=True)
|
||||
basepath = os.path.basename(curfile)
|
||||
imname, ext = os.path.splitext(basepath)
|
||||
print('imname',imname)
|
||||
|
||||
allfiles = natsorted(glob(os.path.join(curdir, curfile, 'enc' ,'*.csv')))
|
||||
|
||||
# get the experiment data (csv format)
|
||||
for subcsv in allfiles:
|
||||
fixations = pd.read_csv(subcsv, header=None)
|
||||
x=[]
|
||||
y=[]
|
||||
duration=[]
|
||||
for row in fixations.iterrows():
|
||||
x.append(row[1][1])
|
||||
y.append(row[1][2])
|
||||
duration.append(row[1][3])
|
||||
|
||||
gif_image=[]
|
||||
|
||||
# generate spotlight gif image
|
||||
for i in range(len(x)):
|
||||
|
||||
#adjust the duration parameter so that the spotlight circle won't be too large or too small to display
|
||||
dur = duration[i]/3 if duration[i]/3>100 else 100
|
||||
dur = dur if dur>100 else 100
|
||||
|
||||
# Make the display time of each spotlight correspond to the observation time of the experimenter
|
||||
dur_int=int(duration[i]/50)
|
||||
for j in range(dur_int):
|
||||
gif_image.append(highlight_function(int(x[i]),int(y[i]),int(dur),os.path.join(origimdir, curfile+'.png'),outpath,imname+'('+str(i+1)+')') )
|
||||
print(i)
|
||||
|
||||
# save spotlight gif image
|
||||
imageio.mimsave(os.path.join(outpath,imname+'.gif'), gif_image, fps=100)
|
||||
|
||||
# problem!!! : The generated spotlight gif image is too large
|
||||
# possible solution: convert gif format to webm format. Converting to webm format can significantly reduce the file size
|
||||
|
||||
|
|
@ -0,0 +1,89 @@
|
|||
import generate_codecharts
|
||||
import generate_sentinels
|
||||
import os
|
||||
import string
|
||||
import random
|
||||
import json
|
||||
import matplotlib.pyplot as plt
|
||||
import numpy as np
|
||||
import base64
|
||||
import glob
|
||||
|
||||
def generate_tutorials(tutorial_image_dir,rootdir,image_width,image_height,border_padding,N,target_type,target_imdir,N_sent=0):
|
||||
|
||||
if not os.path.exists(tutorial_image_dir):
|
||||
raise Exception('Please include a directory of tutorial images at: %s'%(tutorial_image_dir))
|
||||
|
||||
tutorial_images = []
|
||||
for ext in ('*.jpeg', '*.png', '*.jpg'):
|
||||
tutorial_images.extend(glob.glob(os.path.join(tutorial_image_dir, ext)))
|
||||
print('A total of %d images will be sampled from for the tutorials.'%(len(tutorial_images)))
|
||||
|
||||
tutorial_CC_dir = os.path.join(rootdir,'tutorial_CC')
|
||||
if not os.path.exists(tutorial_CC_dir):
|
||||
os.makedirs(tutorial_CC_dir)
|
||||
|
||||
tutorial_sentinel_dir = os.path.join(rootdir,'tutorial_sentinels')
|
||||
if not os.path.exists(tutorial_sentinel_dir):
|
||||
os.makedirs(tutorial_sentinel_dir)
|
||||
|
||||
# make the corresponding codecharts
|
||||
data = {}
|
||||
data_with_coords = {}
|
||||
for img_num in range(N):
|
||||
filename = tutorial_images[img_num]
|
||||
codechart_filename = os.path.join(tutorial_CC_dir,'tutorial_real_CC_%d.jpg'%(img_num))
|
||||
valid_codes, coordinates = generate_codecharts.create_codechart(codechart_filename,image_width,image_height)
|
||||
data[filename] = {'valid_codes':valid_codes,'flag':'tutorial_real','codechart_file':codechart_filename}
|
||||
data_with_coords[filename] = {'valid_codes':valid_codes, 'coordinates':coordinates,\
|
||||
'flag':'tutorial_real','codechart_file':codechart_filename}
|
||||
|
||||
# now generate sentinel images (also N) with their corresponding codecharts
|
||||
print('Populating %s with %d sentinel images'%(tutorial_sentinel_dir,N_sent))
|
||||
print('Populating %s with %d corresponding codecharts'%(tutorial_CC_dir,N_sent))
|
||||
for img_num in range(N_sent):
|
||||
codechart_filename = os.path.join(tutorial_CC_dir,'tutorial_sentinel_CC_%d.jpg'%(img_num))
|
||||
sentinel_filename = os.path.join(tutorial_sentinel_dir,'tutorial_sentinel_%d.jpg'%(img_num))
|
||||
D,D_full = generate_sentinels.make_sentinel(codechart_filename,sentinel_filename,\
|
||||
image_width,image_height,border_padding,target_type, target_imdir)
|
||||
D['flag'] = 'tutorial_sentinel'
|
||||
D['codechart_file'] = codechart_filename
|
||||
D_full['flag'] = 'tutorial_sentinel'
|
||||
D_full['codechart_file'] = codechart_filename
|
||||
data[sentinel_filename] = D
|
||||
data_with_coords[sentinel_filename] = D_full
|
||||
|
||||
with open(os.path.join(rootdir,'tutorial.json'), 'w') as outfile:
|
||||
json.dump(data, outfile)
|
||||
print('Writing out %s'%(os.path.join(rootdir,'tutorial.json')))
|
||||
|
||||
with open(os.path.join(rootdir,'tutorial_full.json'), 'w') as outfile:
|
||||
json.dump(data_with_coords, outfile)
|
||||
print('Writing out %s'%(os.path.join(rootdir,'tutorial_full.json')))
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
|
||||
image_width = 1920
|
||||
image_height = 1080
|
||||
border_padding = 100 # don't put fixation cross in this region of the image
|
||||
rootdir = './task_data'
|
||||
target_type = "red_dot"
|
||||
target_imdir = ""
|
||||
|
||||
tutorial_image_dir = os.path.join(rootdir,'tutorial_images')
|
||||
|
||||
tutorial_images = glob.glob(os.path.join(tutorial_image_dir,'*.jpg'))
|
||||
N = len(tutorial_images)
|
||||
# assume that tutorial_images contains as many images (N) as desired for the tutorials
|
||||
|
||||
generate_tutorials(tutorial_image_dir,rootdir,image_width,image_height,border_padding,N,target_type,target_imdir)
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
581
WebInterface/Front-end/generate-experiment-files/main.ipynb
Normal file
581
WebInterface/Front-end/generate-experiment-files/main.ipynb
Normal file
|
|
@ -0,0 +1,581 @@
|
|||
{
|
||||
"cells": [
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"## Experimental set-up: ##\n",
|
||||
"\n",
|
||||
"This code will generate experimental files that can either be independently hosted on a website and run with recruited participants, or via our [MTurk iPython notebook](https://github.com/a-newman/mturk-api-notebook) be used for launching Amazon Mechanical Turk (MTurk) HITs. \n",
|
||||
"\n",
|
||||
"An experiment is composed of different sets of images:\n",
|
||||
"* **target images** are the images you want to collect attention data on - those are images that you provide (in directory `sourcedir` below)\n",
|
||||
"* **tutorial images** are images that will be shown to participants at the beginning of the experiment to get them familiarized with the codecharts set-up (you can reuse the tutorial image we provide, or provide your own in directory `tutorial_source_dir` below)\n",
|
||||
" * *hint: if your images are very different in content from the images in our set, you may want to train your participants on your own images, to avoid a context switch between the tutorial and main experiment*\n",
|
||||
"* **sentinel images** are images interspersed throughout the experiment where participant attention is guided to a very specific point on the screen, used as validation/calibration images to ensure participants are actually moving their eyes and looking where they're supposed to; the code below will intersperse images from the `sentinel_target_images` directory we provide throughout your experimental sequence\n",
|
||||
" * sentinel images can be interspersed throughout both the tutorial and target images, or excluded from the tutorial (via `add_sentinels_to_tutorial` flag below); we recommend having sentinel images as part of the tutorial to familiarize participants with such images as well\n",
|
||||
" \n",
|
||||
"The code below will populate the `rootdir` task directory with #`num_subject_files` subject files for you, where each subject file corresponds to an experiment you can run on a single participant. For each subject file, a set of #`num_images_per_sf` will be randomly sampled from the `sourcedir` image directory. A set of #`num_sentinels_per_sf` sentinel images will also be sampled from the `sentinel_imdir` image directory, and distributed throughout the experiment. A tutorial will be generated at the beginning of the experiment with #`num_imgs_per_tutorial` randomly sampled from the `tutorial_source_dir` image directory, along with an additional #`num_sentinels_per_tutorial` sentinel files distributed throughout the tutorial (if `add_sentinels_to_tutorial` flag is set to true). "
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 1,
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"import os\n",
|
||||
"import string\n",
|
||||
"import random\n",
|
||||
"import json \n",
|
||||
"import matplotlib.pyplot as plt\n",
|
||||
"import numpy as np\n",
|
||||
"import base64 \n",
|
||||
"import glob"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 4,
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"sourcedir = '10' # replace this with your own directory of experiment images\n",
|
||||
"filldir = '1' # the filler images for False alarm and Correct rejection\n",
|
||||
"blurdir = '10_blur' # blurry images"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 5,
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"# PARAMETERS for generating subject files\n",
|
||||
"num_subject_files = 1 # number of subject files to generate (i.e., # of mturk assignments that will be put up) \n",
|
||||
"num_images_per_sf = 20 # number of target images per subject file \n",
|
||||
"num_imgs_per_tutorial = 0 # number of tutorial images per subject file\n",
|
||||
"num_sentinels_per_sf = 0 # number of sentinel images to distribute throughout the experiment (excluding the tutorial)\n",
|
||||
"add_sentinels_to_tutorial = False # whether to have sentinel images as part of the tutorial\n",
|
||||
"num_sentinels_per_tutorial = 0 # number of sentinel images to distribute throughout the tutorial"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Another bit of terminology and experimental logistics involves **buckets** which are a way to distribute experiment stimuli so that multiple experiments can be run in parallel (and participants can be reused for different subsets of images). If you have a lot of images that you want to collect data on, and for each participant you are sampling a set of only #`num_images_per_sf`, then you might have to generate a large `num_subject_files` in order to have enough data points per image. A way to speed up data collection is to split all the target images into #`num_buckets` disjoint buckets, and then to generate subject files per bucket. Given that subject files generated per bucket are guaranteed to have a disjoint set of images, the same participant can be run on multiple subject files from different buckets. MTurk HITs corresponding to different buckets can be launched all at once. In summary, in MTurk terms, you can generate as many HITs as `num_buckets` specified below, and as many assignments per HIT as `num_subject_files`. \n",
|
||||
"\n",
|
||||
"The way the codecharts methodology works, a jittered grid of alphanumeric triplets appears after every image presentation (whether it is a target, sentinel, or tutorial image), since a participant will need to indicate where on the preceding image s/he looked, by reporting a triplet. To avoid generating an excessive number of codecharts (that bulks up all the subject files), we can reuse some codecharts across buckets. The way we do this is by pre-generating #`ncodecharts` codecharts, and then randomly sampling from these when generating the individual subject files."
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 6,
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"# we pre-generate some codecharts and sentinel images so that we can reuse these across participants and buckets \n",
|
||||
"# and potentially not have to generate as many files; these can be set to any number, and the corresponding code\n",
|
||||
"# will just sample as many images as need per subject file\n",
|
||||
"\n",
|
||||
"ncodecharts = num_subject_files*num_images_per_sf # number of codecharts to generate; can be changed\n",
|
||||
"sentinel_images_per_bucket = num_subject_files*num_sentinels_per_sf # can be changed\n",
|
||||
"\n",
|
||||
"# set these parameters\n",
|
||||
"num_buckets = 1 # number of disjoint sets of subject files to create (for running multiple parallel HITs)\n",
|
||||
"start_bucket_at = 0 # you can use this and the next parameter to generate more buckets if running the code later\n",
|
||||
"which_buckets = [0] # a list of specific buckets e.g., [4,5,6] to generate experiment data for\n",
|
||||
"\n",
|
||||
"rootdir = '../assets/task_data' # where all the experiment data will be stored\n",
|
||||
"if not os.path.exists(rootdir):\n",
|
||||
" print('Creating directory %s'%(rootdir))\n",
|
||||
" os.makedirs(rootdir)\n",
|
||||
"\n",
|
||||
"real_image_dir = os.path.join(rootdir,'real_images') # target images, distributed by buckets\n",
|
||||
" # (shared across buckets)\n",
|
||||
"sentinel_image_dir = os.path.join(rootdir,'sentinel_images') # sentinel images, distributed by buckets\n",
|
||||
"sentinel_CC_dir = os.path.join(rootdir,'sentinel_CC') # codecharts corresponding to the sentinel images\n",
|
||||
" # (shared across buckets)\n",
|
||||
"#sentinel_targetim_dir = os.path.join(rootdir, 'sentinel_target') \n",
|
||||
"real_blurred_dir = os.path.join(rootdir,'real_blurred')\n",
|
||||
"real_filler_dir = os.path.join(rootdir,'real_filler')"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 7,
|
||||
"metadata": {},
|
||||
"outputs": [
|
||||
{
|
||||
"name": "stdout",
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
"20 files copied from 10 to ../assets/task_data/all_images\n",
|
||||
"Padding 20 image files to dimensions: [1734,1340]...\n",
|
||||
"Done!\n"
|
||||
]
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"# this cell creates an `all_images` directory, copies images from sourcedir, and pads them to the required dimensions\n",
|
||||
"\n",
|
||||
"import create_padded_image_dir\n",
|
||||
"\n",
|
||||
"all_image_dir = os.path.join(rootdir,'all_images')\n",
|
||||
"if not os.path.exists(all_image_dir):\n",
|
||||
" print('Creating directory %s'%(all_image_dir))\n",
|
||||
" os.makedirs(all_image_dir)\n",
|
||||
" \n",
|
||||
"allfiles = []\n",
|
||||
"for ext in ('*.jpeg', '*.png', '*.jpg'):\n",
|
||||
" allfiles.extend(glob.glob(os.path.join(sourcedir, ext)))\n",
|
||||
"print(\"%d files copied from %s to %s\"%(len(allfiles),sourcedir,all_image_dir))\n",
|
||||
" \n",
|
||||
"image_width,image_height = create_padded_image_dir.save_padded_images(all_image_dir,allfiles)"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 8,
|
||||
"metadata": {},
|
||||
"outputs": [
|
||||
{
|
||||
"name": "stdout",
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
"20 files copied from 10_blur to ../assets/task_data/blurred_images\n",
|
||||
"Padding 20 image files to dimensions: [1734,1340]...\n",
|
||||
"Done!\n"
|
||||
]
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"blurred_image_dir = os.path.join(rootdir,'blurred_images')\n",
|
||||
"\n",
|
||||
"if not os.path.exists(blurred_image_dir):\n",
|
||||
" print('Creating directory %s'%(blurred_image_dir))\n",
|
||||
" os.makedirs(blurred_image_dir)\n",
|
||||
" \n",
|
||||
"blurfiles = []\n",
|
||||
"for ext in ('*.jpeg', '*.png', '*.jpg'):\n",
|
||||
" blurfiles.extend(glob.glob(os.path.join(blurdir, ext)))\n",
|
||||
"print(\"%d files copied from %s to %s\"%(len(blurfiles),blurdir,blurred_image_dir))\n",
|
||||
" \n",
|
||||
"image_width,image_height = create_padded_image_dir.save_padded_images(blurred_image_dir,blurfiles)"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 9,
|
||||
"metadata": {},
|
||||
"outputs": [
|
||||
{
|
||||
"name": "stdout",
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
"20 files copied from 1 to ../assets/task_data/filler_images\n",
|
||||
"Padding 20 image files to dimensions: [1786,1340]...\n",
|
||||
"Done!\n"
|
||||
]
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"filler_image_dir = os.path.join(rootdir,'filler_images')\n",
|
||||
"\n",
|
||||
"if not os.path.exists(filler_image_dir):\n",
|
||||
" print('Creating directory %s'%(filler_image_dir))\n",
|
||||
" os.makedirs(filler_image_dir)\n",
|
||||
" \n",
|
||||
"fillerfiles = []\n",
|
||||
"for ext in ('*.jpeg', '*.png', '*.jpg'):\n",
|
||||
" fillerfiles.extend(glob.glob(os.path.join(filldir, ext)))\n",
|
||||
"print(\"%d files copied from %s to %s\"%(len(fillerfiles),filldir,filler_image_dir))\n",
|
||||
" \n",
|
||||
"image_width,image_height = create_padded_image_dir.save_padded_images(filler_image_dir,fillerfiles)"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 10,
|
||||
"metadata": {},
|
||||
"outputs": [
|
||||
{
|
||||
"name": "stdout",
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
"Distributing images across 1 buckets\n",
|
||||
"Populating ../assets/task_data/real_images/bucket0 with 20 images\n"
|
||||
]
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"# this cell creates the requested number of buckets and distributes images from `all_image_dir` to the corresponding\n",
|
||||
"# bucket directories inside `real_image_dir`\n",
|
||||
"\n",
|
||||
"from distribute_image_files_by_buckets import distribute_images\n",
|
||||
"\n",
|
||||
"distribute_images(all_image_dir,real_image_dir,num_buckets,start_bucket_at)"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 11,
|
||||
"metadata": {},
|
||||
"outputs": [
|
||||
{
|
||||
"name": "stdout",
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
"Distributing images across 1 buckets\n",
|
||||
"Populating ../assets/task_data/real_blurred/bucket0 with 20 images\n"
|
||||
]
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"distribute_images(blurred_image_dir,real_blurred_dir,num_buckets,start_bucket_at)"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 12,
|
||||
"metadata": {},
|
||||
"outputs": [
|
||||
{
|
||||
"name": "stdout",
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
"Distributing images across 1 buckets\n",
|
||||
"Populating ../assets/task_data/real_filler/bucket0 with 20 images\n"
|
||||
]
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"distribute_images(filler_image_dir,real_filler_dir,num_buckets,start_bucket_at)"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"We create sentinel images by taking a small object (one of a: fixation cross, red dot, or image of a face) and choosing a random location for it on a blank image (away from the image boundaries by at least `border_padding` pixels). The code below creates #`sentinel_images_per_bucket` such sentinel images in each bucket. "
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Now that all the previous cells have generated the requisite image, codechart, sentinel, and tutorial files, the following code will generate `num_subject_files` individual subject files by sampling from the appropriate image directories and creating an experimental sequence. "
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"# Only run this if you want new subject json"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 11,
|
||||
"metadata": {},
|
||||
"outputs": [
|
||||
{
|
||||
"name": "stdout",
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
"['../assets/task_data/real_blurred/bucket0/wsj265.png', '../assets/task_data/real_blurred/bucket0/treasuryD07_3.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_242.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_257.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_243.png', '../assets/task_data/real_blurred/bucket0/whoJ43_1.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_150.png', '../assets/task_data/real_blurred/bucket0/3iRYXLvZ8oVQDMLR-CebnQ==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_194.png', '../assets/task_data/real_blurred/bucket0/whoQ12_2.png', '../assets/task_data/real_blurred/bucket0/whoF03.png', '../assets/task_data/real_blurred/bucket0/wsj3.png', '../assets/task_data/real_blurred/bucket0/whoB10_1.png', '../assets/task_data/real_blurred/bucket0/3LY3OX8bU7uKhgcRPgDRxw==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_262.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_260.png', '../assets/task_data/real_blurred/bucket0/0BmWZbQdEukHi79Lit01oQ==.0.png', '../assets/task_data/real_blurred/bucket0/wsj86.png', '../assets/task_data/real_blurred/bucket0/whoJ36_2.png', '../assets/task_data/real_blurred/bucket0/whoK21.png']\n",
|
||||
"../assets/task_data/real_blurred/bucket0/*.jpg\n",
|
||||
"Generating 1 subject files in bucket 0\n",
|
||||
"['../assets/task_data/real_blurred/bucket0/wsj265.png', '../assets/task_data/real_blurred/bucket0/treasuryD07_3.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_242.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_257.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_243.png', '../assets/task_data/real_blurred/bucket0/whoJ43_1.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_150.png', '../assets/task_data/real_blurred/bucket0/3iRYXLvZ8oVQDMLR-CebnQ==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_194.png', '../assets/task_data/real_blurred/bucket0/whoQ12_2.png', '../assets/task_data/real_blurred/bucket0/whoF03.png', '../assets/task_data/real_blurred/bucket0/wsj3.png', '../assets/task_data/real_blurred/bucket0/whoB10_1.png', '../assets/task_data/real_blurred/bucket0/3LY3OX8bU7uKhgcRPgDRxw==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_262.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_260.png', '../assets/task_data/real_blurred/bucket0/0BmWZbQdEukHi79Lit01oQ==.0.png', '../assets/task_data/real_blurred/bucket0/wsj86.png', '../assets/task_data/real_blurred/bucket0/whoJ36_2.png', '../assets/task_data/real_blurred/bucket0/whoK21.png']\n",
|
||||
"wsj265.json\n",
|
||||
"['../assets/task_data/real_blurred/bucket0/wsj265.png', '../assets/task_data/real_blurred/bucket0/treasuryD07_3.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_242.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_257.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_243.png', '../assets/task_data/real_blurred/bucket0/whoJ43_1.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_150.png', '../assets/task_data/real_blurred/bucket0/3iRYXLvZ8oVQDMLR-CebnQ==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_194.png', '../assets/task_data/real_blurred/bucket0/whoQ12_2.png', '../assets/task_data/real_blurred/bucket0/whoF03.png', '../assets/task_data/real_blurred/bucket0/wsj3.png', '../assets/task_data/real_blurred/bucket0/whoB10_1.png', '../assets/task_data/real_blurred/bucket0/3LY3OX8bU7uKhgcRPgDRxw==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_262.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_260.png', '../assets/task_data/real_blurred/bucket0/0BmWZbQdEukHi79Lit01oQ==.0.png', '../assets/task_data/real_blurred/bucket0/wsj86.png', '../assets/task_data/real_blurred/bucket0/whoJ36_2.png', '../assets/task_data/real_blurred/bucket0/whoK21.png']\n",
|
||||
"treasuryD07_3.json\n",
|
||||
"['../assets/task_data/real_blurred/bucket0/wsj265.png', '../assets/task_data/real_blurred/bucket0/treasuryD07_3.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_242.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_257.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_243.png', '../assets/task_data/real_blurred/bucket0/whoJ43_1.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_150.png', '../assets/task_data/real_blurred/bucket0/3iRYXLvZ8oVQDMLR-CebnQ==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_194.png', '../assets/task_data/real_blurred/bucket0/whoQ12_2.png', '../assets/task_data/real_blurred/bucket0/whoF03.png', '../assets/task_data/real_blurred/bucket0/wsj3.png', '../assets/task_data/real_blurred/bucket0/whoB10_1.png', '../assets/task_data/real_blurred/bucket0/3LY3OX8bU7uKhgcRPgDRxw==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_262.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_260.png', '../assets/task_data/real_blurred/bucket0/0BmWZbQdEukHi79Lit01oQ==.0.png', '../assets/task_data/real_blurred/bucket0/wsj86.png', '../assets/task_data/real_blurred/bucket0/whoJ36_2.png', '../assets/task_data/real_blurred/bucket0/whoK21.png']\n",
|
||||
"economist_daily_chart_242.json\n",
|
||||
"['../assets/task_data/real_blurred/bucket0/wsj265.png', '../assets/task_data/real_blurred/bucket0/treasuryD07_3.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_242.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_257.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_243.png', '../assets/task_data/real_blurred/bucket0/whoJ43_1.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_150.png', '../assets/task_data/real_blurred/bucket0/3iRYXLvZ8oVQDMLR-CebnQ==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_194.png', '../assets/task_data/real_blurred/bucket0/whoQ12_2.png', '../assets/task_data/real_blurred/bucket0/whoF03.png', '../assets/task_data/real_blurred/bucket0/wsj3.png', '../assets/task_data/real_blurred/bucket0/whoB10_1.png', '../assets/task_data/real_blurred/bucket0/3LY3OX8bU7uKhgcRPgDRxw==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_262.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_260.png', '../assets/task_data/real_blurred/bucket0/0BmWZbQdEukHi79Lit01oQ==.0.png', '../assets/task_data/real_blurred/bucket0/wsj86.png', '../assets/task_data/real_blurred/bucket0/whoJ36_2.png', '../assets/task_data/real_blurred/bucket0/whoK21.png']\n",
|
||||
"economist_daily_chart_257.json\n",
|
||||
"['../assets/task_data/real_blurred/bucket0/wsj265.png', '../assets/task_data/real_blurred/bucket0/treasuryD07_3.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_242.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_257.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_243.png', '../assets/task_data/real_blurred/bucket0/whoJ43_1.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_150.png', '../assets/task_data/real_blurred/bucket0/3iRYXLvZ8oVQDMLR-CebnQ==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_194.png', '../assets/task_data/real_blurred/bucket0/whoQ12_2.png', '../assets/task_data/real_blurred/bucket0/whoF03.png', '../assets/task_data/real_blurred/bucket0/wsj3.png', '../assets/task_data/real_blurred/bucket0/whoB10_1.png', '../assets/task_data/real_blurred/bucket0/3LY3OX8bU7uKhgcRPgDRxw==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_262.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_260.png', '../assets/task_data/real_blurred/bucket0/0BmWZbQdEukHi79Lit01oQ==.0.png', '../assets/task_data/real_blurred/bucket0/wsj86.png', '../assets/task_data/real_blurred/bucket0/whoJ36_2.png', '../assets/task_data/real_blurred/bucket0/whoK21.png']\n",
|
||||
"economist_daily_chart_243.json\n",
|
||||
"['../assets/task_data/real_blurred/bucket0/wsj265.png', '../assets/task_data/real_blurred/bucket0/treasuryD07_3.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_242.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_257.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_243.png', '../assets/task_data/real_blurred/bucket0/whoJ43_1.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_150.png', '../assets/task_data/real_blurred/bucket0/3iRYXLvZ8oVQDMLR-CebnQ==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_194.png', '../assets/task_data/real_blurred/bucket0/whoQ12_2.png', '../assets/task_data/real_blurred/bucket0/whoF03.png', '../assets/task_data/real_blurred/bucket0/wsj3.png', '../assets/task_data/real_blurred/bucket0/whoB10_1.png', '../assets/task_data/real_blurred/bucket0/3LY3OX8bU7uKhgcRPgDRxw==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_262.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_260.png', '../assets/task_data/real_blurred/bucket0/0BmWZbQdEukHi79Lit01oQ==.0.png', '../assets/task_data/real_blurred/bucket0/wsj86.png', '../assets/task_data/real_blurred/bucket0/whoJ36_2.png', '../assets/task_data/real_blurred/bucket0/whoK21.png']\n",
|
||||
"whoJ43_1.json\n",
|
||||
"['../assets/task_data/real_blurred/bucket0/wsj265.png', '../assets/task_data/real_blurred/bucket0/treasuryD07_3.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_242.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_257.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_243.png', '../assets/task_data/real_blurred/bucket0/whoJ43_1.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_150.png', '../assets/task_data/real_blurred/bucket0/3iRYXLvZ8oVQDMLR-CebnQ==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_194.png', '../assets/task_data/real_blurred/bucket0/whoQ12_2.png', '../assets/task_data/real_blurred/bucket0/whoF03.png', '../assets/task_data/real_blurred/bucket0/wsj3.png', '../assets/task_data/real_blurred/bucket0/whoB10_1.png', '../assets/task_data/real_blurred/bucket0/3LY3OX8bU7uKhgcRPgDRxw==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_262.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_260.png', '../assets/task_data/real_blurred/bucket0/0BmWZbQdEukHi79Lit01oQ==.0.png', '../assets/task_data/real_blurred/bucket0/wsj86.png', '../assets/task_data/real_blurred/bucket0/whoJ36_2.png', '../assets/task_data/real_blurred/bucket0/whoK21.png']\n",
|
||||
"economist_daily_chart_150.json\n",
|
||||
"['../assets/task_data/real_blurred/bucket0/wsj265.png', '../assets/task_data/real_blurred/bucket0/treasuryD07_3.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_242.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_257.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_243.png', '../assets/task_data/real_blurred/bucket0/whoJ43_1.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_150.png', '../assets/task_data/real_blurred/bucket0/3iRYXLvZ8oVQDMLR-CebnQ==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_194.png', '../assets/task_data/real_blurred/bucket0/whoQ12_2.png', '../assets/task_data/real_blurred/bucket0/whoF03.png', '../assets/task_data/real_blurred/bucket0/wsj3.png', '../assets/task_data/real_blurred/bucket0/whoB10_1.png', '../assets/task_data/real_blurred/bucket0/3LY3OX8bU7uKhgcRPgDRxw==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_262.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_260.png', '../assets/task_data/real_blurred/bucket0/0BmWZbQdEukHi79Lit01oQ==.0.png', '../assets/task_data/real_blurred/bucket0/wsj86.png', '../assets/task_data/real_blurred/bucket0/whoJ36_2.png', '../assets/task_data/real_blurred/bucket0/whoK21.png']\n",
|
||||
"3iRYXLvZ8oVQDMLR-CebnQ==.0.json\n",
|
||||
"['../assets/task_data/real_blurred/bucket0/wsj265.png', '../assets/task_data/real_blurred/bucket0/treasuryD07_3.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_242.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_257.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_243.png', '../assets/task_data/real_blurred/bucket0/whoJ43_1.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_150.png', '../assets/task_data/real_blurred/bucket0/3iRYXLvZ8oVQDMLR-CebnQ==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_194.png', '../assets/task_data/real_blurred/bucket0/whoQ12_2.png', '../assets/task_data/real_blurred/bucket0/whoF03.png', '../assets/task_data/real_blurred/bucket0/wsj3.png', '../assets/task_data/real_blurred/bucket0/whoB10_1.png', '../assets/task_data/real_blurred/bucket0/3LY3OX8bU7uKhgcRPgDRxw==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_262.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_260.png', '../assets/task_data/real_blurred/bucket0/0BmWZbQdEukHi79Lit01oQ==.0.png', '../assets/task_data/real_blurred/bucket0/wsj86.png', '../assets/task_data/real_blurred/bucket0/whoJ36_2.png', '../assets/task_data/real_blurred/bucket0/whoK21.png']\n",
|
||||
"economist_daily_chart_194.json\n",
|
||||
"['../assets/task_data/real_blurred/bucket0/wsj265.png', '../assets/task_data/real_blurred/bucket0/treasuryD07_3.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_242.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_257.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_243.png', '../assets/task_data/real_blurred/bucket0/whoJ43_1.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_150.png', '../assets/task_data/real_blurred/bucket0/3iRYXLvZ8oVQDMLR-CebnQ==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_194.png', '../assets/task_data/real_blurred/bucket0/whoQ12_2.png', '../assets/task_data/real_blurred/bucket0/whoF03.png', '../assets/task_data/real_blurred/bucket0/wsj3.png', '../assets/task_data/real_blurred/bucket0/whoB10_1.png', '../assets/task_data/real_blurred/bucket0/3LY3OX8bU7uKhgcRPgDRxw==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_262.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_260.png', '../assets/task_data/real_blurred/bucket0/0BmWZbQdEukHi79Lit01oQ==.0.png', '../assets/task_data/real_blurred/bucket0/wsj86.png', '../assets/task_data/real_blurred/bucket0/whoJ36_2.png', '../assets/task_data/real_blurred/bucket0/whoK21.png']\n",
|
||||
"whoQ12_2.json\n",
|
||||
"['../assets/task_data/real_blurred/bucket0/wsj265.png', '../assets/task_data/real_blurred/bucket0/treasuryD07_3.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_242.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_257.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_243.png', '../assets/task_data/real_blurred/bucket0/whoJ43_1.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_150.png', '../assets/task_data/real_blurred/bucket0/3iRYXLvZ8oVQDMLR-CebnQ==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_194.png', '../assets/task_data/real_blurred/bucket0/whoQ12_2.png', '../assets/task_data/real_blurred/bucket0/whoF03.png', '../assets/task_data/real_blurred/bucket0/wsj3.png', '../assets/task_data/real_blurred/bucket0/whoB10_1.png', '../assets/task_data/real_blurred/bucket0/3LY3OX8bU7uKhgcRPgDRxw==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_262.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_260.png', '../assets/task_data/real_blurred/bucket0/0BmWZbQdEukHi79Lit01oQ==.0.png', '../assets/task_data/real_blurred/bucket0/wsj86.png', '../assets/task_data/real_blurred/bucket0/whoJ36_2.png', '../assets/task_data/real_blurred/bucket0/whoK21.png']\n",
|
||||
"whoF03.json\n",
|
||||
"['../assets/task_data/real_blurred/bucket0/wsj265.png', '../assets/task_data/real_blurred/bucket0/treasuryD07_3.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_242.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_257.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_243.png', '../assets/task_data/real_blurred/bucket0/whoJ43_1.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_150.png', '../assets/task_data/real_blurred/bucket0/3iRYXLvZ8oVQDMLR-CebnQ==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_194.png', '../assets/task_data/real_blurred/bucket0/whoQ12_2.png', '../assets/task_data/real_blurred/bucket0/whoF03.png', '../assets/task_data/real_blurred/bucket0/wsj3.png', '../assets/task_data/real_blurred/bucket0/whoB10_1.png', '../assets/task_data/real_blurred/bucket0/3LY3OX8bU7uKhgcRPgDRxw==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_262.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_260.png', '../assets/task_data/real_blurred/bucket0/0BmWZbQdEukHi79Lit01oQ==.0.png', '../assets/task_data/real_blurred/bucket0/wsj86.png', '../assets/task_data/real_blurred/bucket0/whoJ36_2.png', '../assets/task_data/real_blurred/bucket0/whoK21.png']\n",
|
||||
"wsj3.json\n",
|
||||
"['../assets/task_data/real_blurred/bucket0/wsj265.png', '../assets/task_data/real_blurred/bucket0/treasuryD07_3.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_242.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_257.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_243.png', '../assets/task_data/real_blurred/bucket0/whoJ43_1.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_150.png', '../assets/task_data/real_blurred/bucket0/3iRYXLvZ8oVQDMLR-CebnQ==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_194.png', '../assets/task_data/real_blurred/bucket0/whoQ12_2.png', '../assets/task_data/real_blurred/bucket0/whoF03.png', '../assets/task_data/real_blurred/bucket0/wsj3.png', '../assets/task_data/real_blurred/bucket0/whoB10_1.png', '../assets/task_data/real_blurred/bucket0/3LY3OX8bU7uKhgcRPgDRxw==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_262.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_260.png', '../assets/task_data/real_blurred/bucket0/0BmWZbQdEukHi79Lit01oQ==.0.png', '../assets/task_data/real_blurred/bucket0/wsj86.png', '../assets/task_data/real_blurred/bucket0/whoJ36_2.png', '../assets/task_data/real_blurred/bucket0/whoK21.png']\n",
|
||||
"whoB10_1.json\n",
|
||||
"['../assets/task_data/real_blurred/bucket0/wsj265.png', '../assets/task_data/real_blurred/bucket0/treasuryD07_3.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_242.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_257.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_243.png', '../assets/task_data/real_blurred/bucket0/whoJ43_1.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_150.png', '../assets/task_data/real_blurred/bucket0/3iRYXLvZ8oVQDMLR-CebnQ==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_194.png', '../assets/task_data/real_blurred/bucket0/whoQ12_2.png', '../assets/task_data/real_blurred/bucket0/whoF03.png', '../assets/task_data/real_blurred/bucket0/wsj3.png', '../assets/task_data/real_blurred/bucket0/whoB10_1.png', '../assets/task_data/real_blurred/bucket0/3LY3OX8bU7uKhgcRPgDRxw==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_262.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_260.png', '../assets/task_data/real_blurred/bucket0/0BmWZbQdEukHi79Lit01oQ==.0.png', '../assets/task_data/real_blurred/bucket0/wsj86.png', '../assets/task_data/real_blurred/bucket0/whoJ36_2.png', '../assets/task_data/real_blurred/bucket0/whoK21.png']\n",
|
||||
"3LY3OX8bU7uKhgcRPgDRxw==.0.json\n",
|
||||
"['../assets/task_data/real_blurred/bucket0/wsj265.png', '../assets/task_data/real_blurred/bucket0/treasuryD07_3.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_242.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_257.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_243.png', '../assets/task_data/real_blurred/bucket0/whoJ43_1.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_150.png', '../assets/task_data/real_blurred/bucket0/3iRYXLvZ8oVQDMLR-CebnQ==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_194.png', '../assets/task_data/real_blurred/bucket0/whoQ12_2.png', '../assets/task_data/real_blurred/bucket0/whoF03.png', '../assets/task_data/real_blurred/bucket0/wsj3.png', '../assets/task_data/real_blurred/bucket0/whoB10_1.png', '../assets/task_data/real_blurred/bucket0/3LY3OX8bU7uKhgcRPgDRxw==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_262.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_260.png', '../assets/task_data/real_blurred/bucket0/0BmWZbQdEukHi79Lit01oQ==.0.png', '../assets/task_data/real_blurred/bucket0/wsj86.png', '../assets/task_data/real_blurred/bucket0/whoJ36_2.png', '../assets/task_data/real_blurred/bucket0/whoK21.png']\n",
|
||||
"economist_daily_chart_262.json\n",
|
||||
"['../assets/task_data/real_blurred/bucket0/wsj265.png', '../assets/task_data/real_blurred/bucket0/treasuryD07_3.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_242.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_257.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_243.png', '../assets/task_data/real_blurred/bucket0/whoJ43_1.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_150.png', '../assets/task_data/real_blurred/bucket0/3iRYXLvZ8oVQDMLR-CebnQ==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_194.png', '../assets/task_data/real_blurred/bucket0/whoQ12_2.png', '../assets/task_data/real_blurred/bucket0/whoF03.png', '../assets/task_data/real_blurred/bucket0/wsj3.png', '../assets/task_data/real_blurred/bucket0/whoB10_1.png', '../assets/task_data/real_blurred/bucket0/3LY3OX8bU7uKhgcRPgDRxw==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_262.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_260.png', '../assets/task_data/real_blurred/bucket0/0BmWZbQdEukHi79Lit01oQ==.0.png', '../assets/task_data/real_blurred/bucket0/wsj86.png', '../assets/task_data/real_blurred/bucket0/whoJ36_2.png', '../assets/task_data/real_blurred/bucket0/whoK21.png']\n",
|
||||
"economist_daily_chart_260.json\n",
|
||||
"['../assets/task_data/real_blurred/bucket0/wsj265.png', '../assets/task_data/real_blurred/bucket0/treasuryD07_3.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_242.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_257.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_243.png', '../assets/task_data/real_blurred/bucket0/whoJ43_1.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_150.png', '../assets/task_data/real_blurred/bucket0/3iRYXLvZ8oVQDMLR-CebnQ==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_194.png', '../assets/task_data/real_blurred/bucket0/whoQ12_2.png', '../assets/task_data/real_blurred/bucket0/whoF03.png', '../assets/task_data/real_blurred/bucket0/wsj3.png', '../assets/task_data/real_blurred/bucket0/whoB10_1.png', '../assets/task_data/real_blurred/bucket0/3LY3OX8bU7uKhgcRPgDRxw==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_262.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_260.png', '../assets/task_data/real_blurred/bucket0/0BmWZbQdEukHi79Lit01oQ==.0.png', '../assets/task_data/real_blurred/bucket0/wsj86.png', '../assets/task_data/real_blurred/bucket0/whoJ36_2.png', '../assets/task_data/real_blurred/bucket0/whoK21.png']\n",
|
||||
"0BmWZbQdEukHi79Lit01oQ==.0.json\n",
|
||||
"['../assets/task_data/real_blurred/bucket0/wsj265.png', '../assets/task_data/real_blurred/bucket0/treasuryD07_3.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_242.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_257.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_243.png', '../assets/task_data/real_blurred/bucket0/whoJ43_1.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_150.png', '../assets/task_data/real_blurred/bucket0/3iRYXLvZ8oVQDMLR-CebnQ==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_194.png', '../assets/task_data/real_blurred/bucket0/whoQ12_2.png', '../assets/task_data/real_blurred/bucket0/whoF03.png', '../assets/task_data/real_blurred/bucket0/wsj3.png', '../assets/task_data/real_blurred/bucket0/whoB10_1.png', '../assets/task_data/real_blurred/bucket0/3LY3OX8bU7uKhgcRPgDRxw==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_262.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_260.png', '../assets/task_data/real_blurred/bucket0/0BmWZbQdEukHi79Lit01oQ==.0.png', '../assets/task_data/real_blurred/bucket0/wsj86.png', '../assets/task_data/real_blurred/bucket0/whoJ36_2.png', '../assets/task_data/real_blurred/bucket0/whoK21.png']\n",
|
||||
"wsj86.json\n",
|
||||
"['../assets/task_data/real_blurred/bucket0/wsj265.png', '../assets/task_data/real_blurred/bucket0/treasuryD07_3.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_242.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_257.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_243.png', '../assets/task_data/real_blurred/bucket0/whoJ43_1.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_150.png', '../assets/task_data/real_blurred/bucket0/3iRYXLvZ8oVQDMLR-CebnQ==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_194.png', '../assets/task_data/real_blurred/bucket0/whoQ12_2.png', '../assets/task_data/real_blurred/bucket0/whoF03.png', '../assets/task_data/real_blurred/bucket0/wsj3.png', '../assets/task_data/real_blurred/bucket0/whoB10_1.png', '../assets/task_data/real_blurred/bucket0/3LY3OX8bU7uKhgcRPgDRxw==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_262.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_260.png', '../assets/task_data/real_blurred/bucket0/0BmWZbQdEukHi79Lit01oQ==.0.png', '../assets/task_data/real_blurred/bucket0/wsj86.png', '../assets/task_data/real_blurred/bucket0/whoJ36_2.png', '../assets/task_data/real_blurred/bucket0/whoK21.png']\n",
|
||||
"whoJ36_2.json\n",
|
||||
"['../assets/task_data/real_blurred/bucket0/wsj265.png', '../assets/task_data/real_blurred/bucket0/treasuryD07_3.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_242.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_257.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_243.png', '../assets/task_data/real_blurred/bucket0/whoJ43_1.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_150.png', '../assets/task_data/real_blurred/bucket0/3iRYXLvZ8oVQDMLR-CebnQ==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_194.png', '../assets/task_data/real_blurred/bucket0/whoQ12_2.png', '../assets/task_data/real_blurred/bucket0/whoF03.png', '../assets/task_data/real_blurred/bucket0/wsj3.png', '../assets/task_data/real_blurred/bucket0/whoB10_1.png', '../assets/task_data/real_blurred/bucket0/3LY3OX8bU7uKhgcRPgDRxw==.0.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_262.png', '../assets/task_data/real_blurred/bucket0/economist_daily_chart_260.png', '../assets/task_data/real_blurred/bucket0/0BmWZbQdEukHi79Lit01oQ==.0.png', '../assets/task_data/real_blurred/bucket0/wsj86.png', '../assets/task_data/real_blurred/bucket0/whoJ36_2.png', '../assets/task_data/real_blurred/bucket0/whoK21.png']\n",
|
||||
"whoK21.json\n",
|
||||
"Subject file ../assets/task_data/subject_files/bucket0/subject_file_0.json DONE\n"
|
||||
]
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"gt_answers = []\n",
|
||||
"reco_answers = []\n",
|
||||
"start_subjects_at = 0 # where to start creating subject files at (if had created other subject files previously)\n",
|
||||
"#if os.path.exists(os.path.join(rootdir,'subject_files/bucket0')):\n",
|
||||
"# subjfiles = glob.glob(os.path.join(rootdir,'subject_files/bucket0/*.json'))\n",
|
||||
"# start_subjects_at = len(subjfiles)\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"## GENERATING SUBJECT FILES \n",
|
||||
"subjdir = os.path.join(rootdir,'subject_files')\n",
|
||||
"if not os.path.exists(subjdir):\n",
|
||||
" os.makedirs(subjdir)\n",
|
||||
" #os.makedirs(os.path.join(rootdir,'full_subject_files'))\n",
|
||||
"\n",
|
||||
"\n",
|
||||
" \n",
|
||||
"# iterate over all buckets \n",
|
||||
"for b in range(len(which_buckets)): \n",
|
||||
"\n",
|
||||
" bucket = 'bucket%d'%(which_buckets[b])\n",
|
||||
" img_bucket_dir = os.path.join(real_image_dir,bucket)\n",
|
||||
" blur_bucket_dir = os.path.join(real_blurred_dir,bucket)\n",
|
||||
" filler_bucket_dir = os.path.join(real_filler_dir,bucket)\n",
|
||||
" img_files = []\n",
|
||||
" blur_files = []\n",
|
||||
" filler_files = []\n",
|
||||
" for ext in ('*.jpeg', '*.png', '*.jpg'):\n",
|
||||
" img_files.extend(glob.glob(os.path.join(img_bucket_dir, ext)))\n",
|
||||
" \n",
|
||||
" for ext in ('*.jpeg', '*.png', '*.jpg'):\n",
|
||||
" blur_files.extend(glob.glob(os.path.join(blur_bucket_dir, ext)))\n",
|
||||
" print(blur_files)\n",
|
||||
" for ext in ('*.jpeg', '*.png', '*.jpg'):\n",
|
||||
" filler_files.extend(glob.glob(os.path.join(filler_bucket_dir, ext)))\n",
|
||||
" \n",
|
||||
" filler_files.extend(img_files) #40 images for recognition task\n",
|
||||
" random.shuffle(filler_files)\n",
|
||||
" \n",
|
||||
" print(os.path.join(blur_bucket_dir, ext))\n",
|
||||
" #sentinel_bucket_dir = os.path.join(sentinel_image_dir,bucket)\n",
|
||||
" #sentinel_files = glob.glob(os.path.join(sentinel_bucket_dir,'*.jpg'))\n",
|
||||
" \n",
|
||||
" #with open(os.path.join(sentinel_bucket_dir,'sentinel_codes_full.json')) as f:\n",
|
||||
" # sentinel_codes_data = json.load(f) # contains mapping of image path to valid codes\n",
|
||||
" \n",
|
||||
" subjdir = os.path.join(rootdir,'subject_files',bucket)\n",
|
||||
" if not os.path.exists(subjdir):\n",
|
||||
" os.makedirs(subjdir)\n",
|
||||
" #os.makedirs(os.path.join(rootdir,'full_subject_files',bucket))\n",
|
||||
" \n",
|
||||
" print('Generating %d subject files in bucket %d'%(num_subject_files,b))\n",
|
||||
" # for each bucket, generate subject files \n",
|
||||
" for i in range(num_subject_files):\n",
|
||||
" \n",
|
||||
" #random.shuffle(img_files)\n",
|
||||
" #random.shuffle(sentinel_files)\n",
|
||||
" #random.shuffle(real_codecharts)\n",
|
||||
"\n",
|
||||
" # for each subject files, add real images \n",
|
||||
" sf_data = []\n",
|
||||
" full_sf_data = []\n",
|
||||
"\n",
|
||||
" \n",
|
||||
" # initialize temporary arrays, because will shuffle real & sentinel tutorial images before adding to\n",
|
||||
" # final subject files\n",
|
||||
" sf_data_temp = []\n",
|
||||
" full_sf_data_temp = []\n",
|
||||
" \n",
|
||||
" \n",
|
||||
" \n",
|
||||
" \n",
|
||||
" # ADDING REAL IMAGES \n",
|
||||
" for j in range(int(num_images_per_sf/2)):\n",
|
||||
" for k in range(2):\n",
|
||||
" image_data = {}\n",
|
||||
" image_data[\"image\"] = img_files[j*2+k] # stores image path \n",
|
||||
"\n",
|
||||
" # select a code chart\n",
|
||||
" #pathname = real_codecharts[j*2+k] # since shuffled, will pick up first set of random codecharts\n",
|
||||
"\n",
|
||||
" #image_data[\"codechart\"] = pathname # stores codechart path \n",
|
||||
" #image_data[\"codes\"] = real_codes_data[pathname]['valid_codes'] # stores valid codes \n",
|
||||
" image_data[\"flag\"] = 'real' # stores flag of whether we have real or sentinel image\n",
|
||||
"\n",
|
||||
" full_image_data = image_data.copy() # identical to image_data but includes a key for coordinates\n",
|
||||
" #full_image_data[\"coordinates\"] = real_codes_data[pathname]['coordinates'] # store locations - (x, y) coordinate of each triplet \n",
|
||||
"\n",
|
||||
" sf_data.append(image_data)\n",
|
||||
" full_sf_data.append(full_image_data)\n",
|
||||
" \n",
|
||||
" for w in range(2):\n",
|
||||
" blur_data = {}\n",
|
||||
" print(blur_files)\n",
|
||||
" blur_data['image'] = blur_files[j*2+w]\n",
|
||||
" # on Windows comment this out\n",
|
||||
" #QA_file_name = blur_files[j*2+w].split('\\\\')[-1][:-4]+'.json'\n",
|
||||
" QA_file_name = blur_files[j*2+w].split('/')[-1][:-4]+'.json'\n",
|
||||
" \n",
|
||||
" with open(os.path.join(sourcedir,QA_file_name)) as f:\n",
|
||||
" print(QA_file_name)\n",
|
||||
" blur_data['QA'] = json.load(f)\n",
|
||||
" #print(blur_data['QA'])\n",
|
||||
" for item in blur_data['QA']:\n",
|
||||
" if 'answer' in blur_data['QA'][item]:\n",
|
||||
" gt_answers.append(blur_data['QA'][item]['answer'])\n",
|
||||
" #pathname = real_codecharts[j*2+w] # since shuffled, will pick up first set of random codecharts\n",
|
||||
"\n",
|
||||
" #blur_data[\"codechart\"] = pathname # stores codechart path \n",
|
||||
" #blur_data[\"codes\"] = real_codes_data[pathname]['valid_codes'] # stores valid codes \n",
|
||||
" blur_data[\"flag\"] = 'blur'\n",
|
||||
" \n",
|
||||
" full_blur_data = blur_data.copy()\n",
|
||||
" #full_blur_data[\"coordinates\"] = real_codes_data[pathname]['coordinates']\n",
|
||||
" \n",
|
||||
" sf_data.append(blur_data)\n",
|
||||
" full_sf_data.append(full_blur_data)\n",
|
||||
" \n",
|
||||
" for img in filler_files:\n",
|
||||
" filler_data = {}\n",
|
||||
" filler_data[\"image\"] = img # stores image path \n",
|
||||
"\n",
|
||||
" # select a code chart\n",
|
||||
" #pathname = real_codecharts[1] # since shuffled, will pick up first set of random codecharts\n",
|
||||
"\n",
|
||||
" #filler_data[\"codechart\"] = pathname # stores codechart path \n",
|
||||
" #filler_data[\"codes\"] = real_codes_data[pathname]['valid_codes'] # stores valid codes \n",
|
||||
" filler_data[\"flag\"] = 'fill' # stores flag of whether we have real or sentinel image\n",
|
||||
" filler_data[\"showed\"] = \"real_images\" in img\n",
|
||||
" if filler_data[\"showed\"] == True:\n",
|
||||
" reco_answers.append('1')\n",
|
||||
" else:\n",
|
||||
" reco_answers.append('2')\n",
|
||||
"\n",
|
||||
" full_filler_data = filler_data.copy() # identical to image_data but includes a key for coordinates\n",
|
||||
" #full_filler_data[\"coordinates\"] = real_codes_data[pathname]['coordinates'] # store locations - (x, y) coordinate of each triplet \n",
|
||||
"\n",
|
||||
" sf_data.append(filler_data)\n",
|
||||
" full_sf_data.append(full_filler_data)\n",
|
||||
" \n",
|
||||
" \n",
|
||||
"\n",
|
||||
" \n",
|
||||
"\n",
|
||||
" # Add an image_id to each subject file entry\n",
|
||||
" image_id = 0 # represents the index of the image in the subject file \n",
|
||||
" for d in range(len(sf_data)): \n",
|
||||
" sf_data[d]['index'] = image_id\n",
|
||||
" full_sf_data[d]['index'] = image_id\n",
|
||||
" image_id+=1\n",
|
||||
"\n",
|
||||
" subj_num = start_subjects_at+i\n",
|
||||
" with open(os.path.join(rootdir,'subject_files',bucket,'subject_file_%d.json'%(subj_num)), 'w') as outfile: \n",
|
||||
" print('Subject file %s DONE'%(outfile.name))\n",
|
||||
" json.dump(sf_data, outfile)\n",
|
||||
" with open(os.path.join(rootdir,'full_subject_files',bucket,'subject_file_%d.json'%(subj_num)), 'w') as outfile: \n",
|
||||
" json.dump(full_sf_data, outfile)"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 16,
|
||||
"metadata": {},
|
||||
"outputs": [
|
||||
{
|
||||
"data": {
|
||||
"text/plain": [
|
||||
"100"
|
||||
]
|
||||
},
|
||||
"execution_count": 16,
|
||||
"metadata": {},
|
||||
"output_type": "execute_result"
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"len(gt_answers)"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 17,
|
||||
"metadata": {},
|
||||
"outputs": [
|
||||
{
|
||||
"data": {
|
||||
"text/plain": [
|
||||
"40"
|
||||
]
|
||||
},
|
||||
"execution_count": 17,
|
||||
"metadata": {},
|
||||
"output_type": "execute_result"
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"len(reco_answers)"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 20,
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"#import numpy as np\n",
|
||||
"#np.save('gt_answers2',gt_answers)\n",
|
||||
"#np.save('recogt_answers2',reco_answers)"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": []
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": []
|
||||
}
|
||||
],
|
||||
"metadata": {
|
||||
"kernelspec": {
|
||||
"display_name": "Python 3 (ipykernel)",
|
||||
"language": "python",
|
||||
"name": "python3"
|
||||
},
|
||||
"language_info": {
|
||||
"codemirror_mode": {
|
||||
"name": "ipython",
|
||||
"version": 3
|
||||
},
|
||||
"file_extension": ".py",
|
||||
"mimetype": "text/x-python",
|
||||
"name": "python",
|
||||
"nbconvert_exporter": "python",
|
||||
"pygments_lexer": "ipython3",
|
||||
"version": "3.8.11"
|
||||
}
|
||||
},
|
||||
"nbformat": 4,
|
||||
"nbformat_minor": 2
|
||||
}
|
||||
Binary file not shown.
|
After Width: | Height: | Size: 8.8 KiB |
Binary file not shown.
|
After Width: | Height: | Size: 7.4 KiB |
Binary file not shown.
|
After Width: | Height: | Size: 7.6 KiB |
Binary file not shown.
|
After Width: | Height: | Size: 9.6 KiB |
Binary file not shown.
|
After Width: | Height: | Size: 8.1 KiB |
Binary file not shown.
|
After Width: | Height: | Size: 6.6 KiB |
Binary file not shown.
|
After Width: | Height: | Size: 7.7 KiB |
Binary file not shown.
|
After Width: | Height: | Size: 7.3 KiB |
Binary file not shown.
|
After Width: | Height: | Size: 7.3 KiB |
Binary file not shown.
|
After Width: | Height: | Size: 8.1 KiB |
Binary file not shown.
|
After Width: | Height: | Size: 144 KiB |
Binary file not shown.
|
After Width: | Height: | Size: 127 KiB |
Binary file not shown.
|
After Width: | Height: | Size: 247 KiB |
Binary file not shown.
|
After Width: | Height: | Size: 213 KiB |
Binary file not shown.
|
After Width: | Height: | Size: 129 KiB |
Binary file not shown.
|
After Width: | Height: | Size: 181 KiB |
Binary file not shown.
|
After Width: | Height: | Size: 124 KiB |
Binary file not shown.
|
After Width: | Height: | Size: 101 KiB |
Binary file not shown.
|
After Width: | Height: | Size: 149 KiB |
Loading…
Add table
Add a link
Reference in a new issue