import os, sys, shutil, random, pathlib
from os.path import join as pjoin
import cv2
import utils
utils.help_option('''
crops_dir2dataset_dir
Convert 'crops_dir' to dataset directory
It make train,valid and test directory in 'crops_dir'
make list of pair of image:mask
shuffle the list and then move image, masks into train/valid/test
NOTE: the names of the masks are changed to the same as the image.
[synopsys]
python crops_dir2dataset_dir.py crops_dir label_str
ex)
python crops_dir2dataset_dir.py 35crops _mask_
''')
crops_dir = sys.argv[1]
crops_dir = pathlib.Path(crops_dir).parts[0]
label_str = sys.argv[2]
all_paths = list(utils.file_paths(crops_dir))
img_paths = sorted(filter(lambda p: label_str not in p, all_paths))
mask_paths = sorted(filter(lambda p: label_str in p, all_paths))
img_mask_pairs = list(zip(img_paths, mask_paths))
random.shuffle(img_mask_pairs)
import utils
utils.help_option('''
mask_maker
make answer data from 'job_records'
and then save them into 'answer_dir'
if you want to start specific 'index' of job records,
(want to rewrite some answers) pass integer as 4th argument.
synopsis
python mask_maker.py job_records answer_dir
python mask_maker.py job_records answer_dir index
example
python mask_maker.py ./job_records.bin ./answers
python mask_maker.py ./job_records.bin ./answers 13
''')
import manual_selector, textMaskMakerUI
import sys, shutil, os, cv2
def is_done(imgpath_imgname):
return (len(imgpath_imgname) == 3)
def main(job_records_path, answer_dir, goto=None):
now_idx, jobs, selected = manual_selector.load(job_records_path)
if goto is not None:
import sys, os, cv2
from fp import pipe, cmap, cfilter
import numpy as np
import utils
np.set_printoptions(threshold=np.nan, linewidth=np.nan)
utils.help_option('''
clean_masks
load masks from 'mask_dir',
clean masks, merge 'selected_channel' to get graysacle image,
and then *OVERWIRTE* masks with clean masks.
it can know dataset is gray or rgb images.
but DO NOT MIX graysacle & rgb images!
[synopsys]
python clean_masks.py mask_dir selected_channel
ex)
python clean_masks.py ./35crops/test/label/ rg
''')
def binarization(img, threshold=100):
#cv2.imshow('nb',img); cv2.waitKey(0)
#_,binarized = cv2.threshold(img, threshold, 255, cv2.THRESH_BINARY)
#print(img[:100,:100])
binarized = (img >= threshold).astype(np.uint8) * 255
#print((img >= threshold)[:100,:100])
#cv2.imshow('b',binarized); cv2.waitKey(0)
return binarized
import os, random, _pickle, cv2
from fp import pipe, cmap
import utils
utils.help_option(
'''
manual_selector
select 'N' images from 'imgs_dir'
and save image name and paths into 'job_records_path'.
'monitor_height' is maximum height size of ui
if 'job_records_path' already exists,
continue with this, append to selected.
synopsis
python manual_selector.py N monitor_height imgs_dir job_records_path
example
python manual_selector.py 2 980 ./mangas/ job_records.bin
'''
)
def load(job_records_path):
with open(job_records_path,'rb') as f:
return _pickle.load(f)
def save(now_idx, jobs, selected, job_records_path):
with open(job_records_path,'wb') as f:
_pickle.dump((now_idx,jobs,selected),f)
import pathlib, shutil, sys, cv2
import numpy as np
import utils
from fp import pipe, cmap, cfilter
utils.help_option('''
separator: separate rgb/grayscale image files.
create new argv[2] directory(tree structure preserved),
separate images in argv[1], and move rgb imgs to new directory.
ex)
python separator.py ./data/examples/ ./data/rgb
^~~~~~~~~~~~~~~~ ^~~~~~~~~~
origin img dir new directory for rgb imgs.
''')
is_grayscale = (lambda img: np.all(img[:, :, 0] == img[:, :, 1]) and np.all(
img[:, :, 1] == img[:, :, 2]))
if __name__ == '__main__':
mixed_imgs_path = sys.argv[1]
rgb_imgs_path = sys.argv[2]
utils.safe_copytree(mixed_imgs_path, rgb_imgs_path,
('*.jpg', '*.jpeg', '*.png'))
f = pipe(utils.file_paths, cmap(lambda path: (cv2.imread(path), path)),
cfilter(lambda img_path: img_path[0] is not None),
cfilter(lambda img_path: not is_grayscale(img_path[0])))
old_parent_dir = pathlib.Path(mixed_imgs_path).parts[-1]
import os, sys, _pickle, cv2, h5py
import utils
utils.help_option('''
manual_classifier:
classify good/bad crop from 'crops.h5'
and then save the result to 'ox_list'
synopsis
python manual_classifier crops.h5 ox_list
example
python manual_classifier ./crops.h5 ox_list
''')
def load(ox_list_path):
with open(ox_list_path, 'rb') as f:
return _pickle.load(f)
def save(now_idx, num_checked, ox_list, ox_list_path):
with open(ox_list_path, 'wb') as f:
_pickle.dump((now_idx, num_checked, ox_list), f)
def look_and_decide(image, window_title='o x 4 6 q'):
while True:
cv2.imshow(window_title, image)
key = cv2.waitKey(1) & 0xFF
if (key == ord('o') or # good crop
key == ord('x') or # bad crop
import sys, os, cv2, tqdm, pathlib
from fp import pipe, cmap, cfilter, flatten
import itertools
import utils
utils.help_option('''
img_cutter
copy_tree structure of 'src_path' to 'dst_path'
then save square crops of images in 'src_path'
[synopsys]
python img_cutter.py crop_size src_path dst_path
ex)
python img_cutter.py 256 ./src ./dst
''')
def hw2start_yxs(origin_yx, img_hw, piece_hw):
org_y, org_x = origin_yx
img_h, img_w = img_hw
piece_h, piece_w = piece_hw
for y in range(org_y, img_h, piece_h):
for x in range(org_x, img_w, piece_w):
yield (y, x)
def hw2not_excess_start_yxs(origin_yx, img_hw, piece_hw):
img_h, img_w = img_hw
piece_h, piece_w = piece_hw
def not_excess(yx):
from skimage.measure import compare_ssim
from fp import pipe, cmap, flip, unzip
from layers import completion_net, discrimination_net
from tester_ui import tester_ui
import utils
utils.help_option(
'''
evaluator:
evaluate all complnets in 'complnet_dir'
using (origin,answer,mask) in 'dataset_dir'.
save mse ratio similarity/error, masked/full ssim as yml.
save mean of scores and list of all scores.
[synopsis]
python evaluator.py complnet_dir dataset_dir img_dir
ex)
python evaluator.py olds/192x_200e/ eval-data/mini_evals ./output/small30_1000results/
'''
)
def mse(A,B):
return ((A-B)**2).mean()
def normalized(uint8img):
return uint8img.astype(np.float32) / 255
def inverse_normalized(float32img):
return (float32img * 255).astype(np.uint8)