def main(args):
out_dir = os.path.join(args.out_dir, args.dataset, args.model,
args.preprocessing, 'rs_{}'.format(args.rs))
# create output directory
os.makedirs(out_dir, exist_ok=True)
util.clear_dir(out_dir)
# create logger
logger = util.get_logger(os.path.join(out_dir, 'log.txt'))
logger.info(args)
logger.info('\ntimestamp: {}'.format(datetime.now()))
# run experiment
experiment(args, logger, out_dir)
def main(args):
# define output directory
out_dir = os.path.join(args.out_dir,
args.dataset,
args.model,
args.preprocessing,
args.surrogate,
args.metric,
'rs_{}'.format(args.rs))
# create output directory and clear any previous contents
os.makedirs(out_dir, exist_ok=True)
util.clear_dir(out_dir)
# create logger
logger = util.get_logger(os.path.join(out_dir, 'log.txt'))
logger.info(args)
logger.info('\ntimestamp: {}'.format(datetime.now()))
# run experiment
experiment(args, out_dir, logger)
def main(args):
# change name of dataset if using a subset of the dataset
if args.train_frac < 1.0 and args.train_frac > 0.0:
dataset = '{}_{}'.format(args.dataset, str(args.train_frac).replace('.', 'p'))
# leave name unchanged
else:
dataset = args.dataset
# change preprocessing if using `feature_path` or `feature_output` with `cb`
if args.model == 'cb' and ('feature_path' in args.method or 'feature_output' in args.method):
args.preprocessing = 'standard'
# Leaf Influence cannot handle cat. features...
elif args.method == 'leaf_influence':
args.preprocessing = 'standard'
# define output directory
out_dir = os.path.join(args.out_dir,
dataset,
args.model,
args.preprocessing,
args.method,
'rs_{}'.format(args.rs))
# create output directory and clear any previous contents
os.makedirs(out_dir, exist_ok=True)
util.clear_dir(out_dir)
# create logger
logger = util.get_logger(os.path.join(out_dir, 'log.txt'))
logger.info(args)
logger.info('\ntimestamp: {}'.format(datetime.now()))
# run experiment
experiment(args, logger, out_dir)
def prepare_directories(self):
self.path = os.path.join(self.root_path, self.experiment['path'])
self.protos_path = os.path.join(self.path, 'protos')
self.solver_path = os.path.join(self.protos_path, 'solver.prototxt')
self.model_path = os.path.join(self.protos_path, 'train_val.prototxt')
self.logs_path = os.path.join(self.path, 'logs')
self.scripts_path = os.path.join(self.path, 'scripts')
if self.replace_mode == 1 and os.path.exists(self.protos_path):
self.prepare_protos()
if self.replace_mode in [0, 1] and os.path.exists(self.path):
self.snapshots_path = self.get_snapshots_directory()
return
for d in [self.protos_path, self.logs_path, self.scripts_path]:
clear_dir(d)
self.prepare_protos()
self.prepare_scripts()
self.snapshots_path = self.get_snapshots_directory()
clear_dir(self.snapshots_path)
def main(args):
# define output directory
out_dir = os.path.join(args.out_dir, args.dataset, args.model,
args.preprocessing, 'rs_{}'.format(args.rs))
# create outut directory and clear any previous contents
os.makedirs(out_dir, exist_ok=True)
util.clear_dir(out_dir)
# create logger
logger = util.get_logger(os.path.join(out_dir, 'log.txt'))
logger.info(args)
logger.info('\ntimestamp: {}'.format(datetime.now()))
# write everything printed to stdout to this log file
logfile, stdout, stderr = util.stdout_stderr_to_log(
os.path.join(out_dir, 'log+.txt'))
# run experiment
experiment(args, logger, out_dir, seed=args.rs)
# restore original stdout and stderr settings
util.reset_stdout_stderr(logfile, stdout, stderr)
def prepare_directories(self):
self.path = os.path.join(self.root_path, self.experiment['path'])
self.protos_path = os.path.join(self.path, 'protos')
self.solver_path = os.path.join(self.protos_path, 'solver.prototxt')
self.model_path = os.path.join(self.protos_path, 'train_val.prototxt')
self.logs_path = os.path.join(self.path, 'logs')
self.scripts_path = os.path.join(self.path, 'scripts')
if self.replace_mode == 1 and os.path.exists(self.protos_path):
self.prepare_protos()
if self.replace_mode in [0, 1] and os.path.exists(self.path):
self.snapshots_path = self.get_snapshots_directory()
return
for d in [self.protos_path, self.logs_path, self.scripts_path]:
clear_dir(d)
self.prepare_protos()
self.prepare_scripts()
self.snapshots_path = self.get_snapshots_directory()
clear_dir(self.snapshots_path)
def main(args):
for i in range(args.n_repeats):
# define output directory
out_dir = os.path.join(args.out_dir, args.dataset, args.model,
'flip_{}'.format(args.flip_frac), args.method,
'rs_{}'.format(args.rs))
# create output directory and clear any previous contents
os.makedirs(out_dir, exist_ok=True)
util.clear_dir(out_dir)
# create logger
logger = util.get_logger(os.path.join(out_dir, 'log.txt'))
logger.info(args)
logger.info('\ntimestamp: {}'.format(datetime.now()))
# run experiment
experiment(args, logger, out_dir)
util.remove_logger(logger)
args.rs += 1
def split_parts_for_image(start_y, preproces_path, out_dir, dir_name,
original_dir, collectData, splitMode):
try:
# 第三步图片
segmented_img = cv2.imread(get_seg_file_name(preproces_path))
# 原始图
original_img = cv2.imread(
get_specified_dir(original_dir, dir_name + ".jpg"))
# 原图宽度
original_width = original_img.shape[1]
original_resize = original_width / 1000
width = segmented_img.shape[1]
height = segmented_img.shape[0]
#定义区域
min_area = 50
max_area = width * height
# BGR=>灰度图
mask = cv2.cvtColor((segmented_img != 0).astype(np.uint8),
cv2.COLOR_BGR2GRAY)
# 部件图
partImages = []
# 边缘部件
part_remove_left = 0
part_remove_right = 0
while True:
# 将mask转化为1维数组
# 返回数组mask中值不为零的元素的下标,
nz = np.nonzero(mask.flatten())[0].flatten()
if len(nz) == 0:
break
nz_i = 0
found_mask = None
found_image = None
while True:
index = nz[nz_i]
seed_x = index % width
# 向下取整
seed_y = index // width
ff_mask = np.zeros((height + 2, width + 2), dtype=np.uint8)
area, _, __, rect = cv2.floodFill(mask,
ff_mask, (seed_x, seed_y),
255,
flags=cv2.FLOODFILL_MASK_ONLY
| cv2.FLOODFILL_FIXED_RANGE)
x = rect[0]
y = rect[1]
w = rect[2]
h = rect[3]
# slicing into found rect
roi_mask = ff_mask[y + 1:y + 1 + h, x + 1:x + 1 + w]
found = False
if min_area < area < max_area:
found_mask = roi_mask
newX = x * 4
newY = (y * 4) + start_y
newW = w * 4
newH = h * 4
# 边界模式，跳出循环
if splitMode == "full":
if newX <= 0:
found_mask = None
part_remove_left += 1
if newW + newX >= original_width:
found_mask = None
part_remove_right += 1
startX = newX - 30
endX = newX + newW + 30
startY = newY - 30
endY = newY + newH + 30
found_image = original_img[startY:endY, startX:endX].copy()
found = True
# clearing found component in the mask
mask[y:y + h, x:x + w][roi_mask != 0] = 0
if found:
break
nz_i += 1
if nz_i >= len(nz):
break
if found_mask is not None:
partImages.append(found_image)
# 如果有多个零件，创建目录保存
# hasmorepart = len(partImages) > 1
# if hasmorepart:
out_dir = get_specified_dir(out_dir, dir_name)
if os.path.exists(out_dir):
clear_dir(out_dir)
else:
create_dir(out_dir)
# 分割信息
if part_remove_left or part_remove_right:
fl = open(out_dir + "/data.json", 'w')
r_data = {
"left_remove": part_remove_left,
"right_remove": part_remove_right
}
fl.write(json.dumps(r_data, ensure_ascii=False, indent=2))
fl.close()
# 输出第二部分
part_index = 0
for part in partImages:
title = os.path.splitext(os.path.split(preproces_path)[1])[0]
file_name = os.path.join("", "%s_%02d.png" % (title, part_index))
# if hasmorepart:
out_file = os.path.join(out_dir,
"%s_%02d.png" % (title, part_index))
# else:
# out_file = os.path.join(out_dir, "%s.png" % (title))
cv2.imwrite(out_file, part)
# 支持数据采集
if collectData:
color_thief = ColorThief(out_file)
dominant_color = color_thief.get_color(quality=1)
h, w = part.shape[:2]
datas = {
"name": file_name,
"w": w,
"h": h,
"area": w * h,
"rgb": dominant_color
}
filePath = out_dir + "/data.json"
if (os.path.exists(filePath)):
fl = open(filePath, 'a')
else:
fl = open(filePath, 'w')
fl.write(json.dumps(datas, ensure_ascii=False, indent=2))
fl.close()
part_index += 1
return dir_name + ".png", True
except Exception as _:
return dir_name + ".png", False