def get_new_boards(indir, json_file):
"""extracts boards from new raw images using coordinates from the json file"""
boards = {}
infiles = listdir_nohidden(indir)
with open(json_file, "r") as file:
lines = file.readlines()
for line in lines: #iterates over all files with masks
x = json.loads(line)
url = x["content"]
filename = url.split("_")[-1]
if filename in infiles: #there exists a hi-res raw version of this image.
img = cv2.imread(os.path.join(indir, filename))
scale = float(img.shape[0])
points = x["annotation"][0]["points"]
points = [[scale * p[0], scale * p[1]] for p in points[:-1]]
points = np.array(points, dtype=np.int32)
board = extract_perspective(img, points, cv_globals.BOARD_SIZE)
board = cv2.cvtColor(board, cv2.COLOR_RGB2GRAY)
board = cv2.resize(board, cv_globals.BOARD_SIZE)
boards[filename] = board
return boards
def rename_images(indir):
"""Give images unique IDs"""
for f in listdir_nohidden(indir):
src = os.path.join(indir, f)
unique_filename = str(uuid.uuid4())
dst = os.path.join(indir, unique_filename + ".JPG")
os.rename(src, dst)
print("renamed: {} to {}".format(src, dst))
def resize_images(indir, outdir):
"""Resizes new raw images"""
for f in listdir_nohidden(indir):
src = os.path.join(indir, f)
image = cv2.imread(src)
resized = cv2.resize(image, cv_globals.INPUT_SIZE, interpolation= cv2.INTER_AREA)
dst = os.path.join(outdir, f)
cv2.imwrite(dst, resized)
print("resized image {}".format(src))
def get_test_generator():
img_filenames = listdir_nohidden(test_data_dir + "raw/")
test_imgs = np.array(
list(
map(lambda x: cv2.imread(test_data_dir + "raw/" + x),
img_filenames)))
for i in range(len(test_imgs)):
yield img_filenames[i], test_imgs[i]
import time
if __name__ == "__main__":
# first store the already labelled examples
already_labelled = []
with open("../data/labels.csv", "r") as csvfile:
reader = csv.reader(csvfile)
for row in reader:
already_labelled.append(row[0])
# next start labelling new examples
with open('../data/labels.csv', 'a') as csvfile:
w = csv.writer(csvfile)
filenames = listdir_nohidden("../data/squares/")
filenames = [f for f in filenames]
labelled_sess = 0
for f in filenames:
if f in already_labelled:
continue
img = cv2.imread("../data/squares/" + f)
plt.imshow(img, cmap="gray")
plt.axis("off")
plt.show()
label = input("Enter label: ")
/path/to/cvroot/weights/modeltype/date/modeltype-{perf}.hdf
where performance is a float between 0 and 1 (1 is good).
"""
import cv_globals
from util import listdir_nohidden
import os
def find_best(models):
best_perf = 0.0
for i in range(len(models)):
model = models[i]
perf = float(model.split("-")[1][:-5])
if perf > best_perf:
best = i
return models[best]
models = ["extractor", "classifier"]
for model in models:
training_runs = listdir_nohidden(cv_globals.weights_dir + model)
for run in training_runs:
models = listdir_nohidden(cv_globals.weights_dir + model + "/" + run)
best = find_best(models)
for model_file in models:
if model_file is not best:
to_delete = cv_globals.weights_dir + model + "/" + run + "/" + model_file
print("Removing: {}".format(to_delete))
os.remove(to_delete)
squares.append(board[i*w:(i+1)*w, j*h:(j+1)*h])
names.append(ranks[j]+files[7-i])
squares = np.array(squares)
squares = squares.reshape(squares.shape[0], 64, 64, 1)
return squares, names
if __name__ == "__main__":
# Extract all squares from all boards
print("Extracting squares...")
#board_dir = "../data/boards/"
#square_dir = "../data/squares/"
(_, board_dir, square_dir) = parse_arguments()
board_filenames = listdir_nohidden(board_dir)
filenames = [f for f in board_filenames]
board_imgs = [cv2.imread(board_dir+f, 0) for f in filenames]
for f, b in zip(filenames, board_imgs):
squares, names = extract_squares(b)
for sq, name in zip(squares, names):
#print("{}{}".format(name, sq.shape))
#cv2.imwrite(square_dir + name + f, sq)
pass
print("\rExtracting squares...DONE")
import cv_globals
import os
from os.path import join
from util import listdir_nohidden
import random
indir = join(cv_globals.CVROOT, "computeroot/user_uploads/squares/")
outdir = join(cv_globals.data_root, "squares/")
train_test_split = 0.2
for dirname in listdir_nohidden(indir):
for filename in listdir_nohidden(join(indir, dirname)):
split = "validation" if random.random(
) < train_test_split else "training"
src = join(indir, dirname, filename)
dst = join(outdir, split, dirname, filename)
print("Moving {} to {}".format(
src.split("/")[-3:],
dst.split("/")[-4:]))
os.rename(src, dst)
def load_image_ids():
filenames = [f[:-4] for f in listdir_nohidden(cv_globals.image_dir)]
return filenames
import os
from util import listdir_nohidden
import cv_globals
import uuid
i = 0
bases = [cv_globals.data_root + "squares/validation", cv_globals.data_root + "squares/training"]
tally = {}
for base in bases:
for dirname in listdir_nohidden(base):
for fname in listdir_nohidden(base + "/" + dirname):
if len(fname) not in tally:
tally[len(fname)] = 0
else:
tally[len(fname)] += 1
if len(fname) != 18:
src = base + "/" + dirname + "/" + fname
uid = str(uuid.uuid1())
dst = os.path.join(base, dirname, uid + ".JPG")
#print("saving long file: {}".format(dst))
#os.rename(src, dst)
print(tally)