def get_image_and_label():
image_label_train = {}
for line in open('../data/train.txt').readlines():
line_idx, pic_path, boxes, labels, img_width, img_height = parse_line(line)
# print(line_idx, pic_path, boxes, labels, img_width, img_height)
# image_name = pic_path.strip().split('/')[-1]
image_label_train[line_idx] = [pic_path, labels]
image_label_val = {}
for line in open('../data/val.txt').readlines():
line_idx, pic_path, boxes, labels, img_width, img_height = parse_line(line)
# print(line_idx, pic_path, boxes, labels, img_width, img_height)
# image_name = pic_path.strip().split('/')[-1]
image_label_train[line_idx+30000] = [pic_path, labels]
image_label_val[line_idx] = [pic_path, labels]
return image_label_train, image_label_val
def generate_rot90_txt(self, annotation_txt, txt_dst):
contents = open(annotation_txt, 'r').readlines()
dstfile = open(txt_dst, 'w')
for line in contents:
pic_path, boxes, labels = parse_line(line, is_str=True)
img = get_img(pic_path)
try:
h, w, _ = img.shape
except:
print('wrong', pic_path)
continue
# rimg = np.rot90(img).copy()
img_dstdir, imgname = os.path.split(pic_path)
savepath = os.path.join(img_dstdir, 'rot90_' + imgname)
aline = savepath
for i in range(len(labels)):
b = boxes[i]
x1, y1, x2, y2 = b[1], w - b[2], b[3], w - b[0]
aline = aline + ' {} {} {} {} {}'.format(
labels[i], x1, y1, x2, y2)
# plot_one_box(img, [b[0], b[1], b[2], b[3]], label='test')
# cv2.imshow('ori', img)
# plot_one_box(rimg, [x1, y1, x2, y2], label='test')
# cv2.imshow('test', rimg)
# cv2.waitKey(0)
# aline = aline + ' ' + str(labels[i]) + ' ' + str(b[0]) + ' ' + str(b[1]) + ' ' + str(b[2]) + ' ' + str(b[3])
dstfile.write(aline)
dstfile.write('\n')
dstfile.close()
self.concat_txts(annotation_txt, txt_dst)
def parse_gt_rec(gt_filename, resize_img_size):
'''
parse and re-organize the gt info.
return:
gt_dict: dict. Each key is a img_id, the value is the gt bboxes in the corresponding img.
'''
global gt_dict
if not gt_dict:
resize_w, resize_h = resize_img_size
with open(gt_filename, 'r') as f:
for line in f:
img_id, pic_path, boxes, labels = parse_line(line)
ori_img_size = cv2.imread(pic_path).shape
ori_w, ori_h = ori_img_size[1], ori_img_size[0]
objects = []
for i in range(len(labels)):
x_min, y_min, x_max, y_max = boxes[i]
label = labels[i]
objects.append([
x_min * resize_w / ori_w, y_min * resize_h / ori_h,
x_max * resize_w / ori_w, y_max * resize_h / ori_h,
label
])
gt_dict[img_id] = objects
return gt_dict
def get_image_and_label(phase):
image_label_dict = {}
if phase == 'train':
for line in open('../data/train.txt').readlines():
line_idx, pic_path, boxes, labels, img_width, img_height = parse_line(
line)
# print(line_idx, pic_path, boxes, labels, img_width, img_height)
# image_name = pic_path.strip().split('/')[-1]
image_label_dict[line_idx] = [pic_path, labels]
else:
for line in open('../data/val.txt').readlines():
line_idx, pic_path, boxes, labels, img_width, img_height = parse_line(
line)
# print(line_idx, pic_path, boxes, labels, img_width, img_height)
# image_name = pic_path.strip().split('/')[-1]
image_label_dict[line_idx] = [pic_path, labels]
return image_label_dict
def get_data_from_web_and_save(data_batch_path, data_train_save_path, data_val_save_path, nfs_mount_path, config, logger, project_name="sjht"):
data_batch_path = data_batch_path.strip()
assert os.path.exists(data_batch_path), "data batch path is not exit:%s"%data_batch_path
# assert os.path.exists(data_train_save_path), "data train save path is not exit"
# assert os.path.exists(data_val_save_path), "data val save path is not exit"
data_train_save = open(data_train_save_path, "w")
data_val_save = open(data_val_save_path, "w")
delete_labels = config.data_set["delete_labels"]
delete_labels_background = config.data_set["delete_labels_background"]
batch_lines = [line.strip() for line in open(data_batch_path).readlines()]
data_train_batch_id = batch_lines.index("#train_data_batch")
data_train_background_batch_id = batch_lines.index("#train_data_background_batch")
val_data_batch_id = batch_lines.index("#val_data_batch")
data_train_batch = batch_lines[data_train_batch_id: data_train_background_batch_id]
data_train_background_batch = batch_lines[data_train_background_batch_id: val_data_batch_id]
data_val_batch = batch_lines[val_data_batch_id:]
data_train = get_data(logger, data_train_batch, project_name=project_name, delete_labels=delete_labels, nfs_mount_path=nfs_mount_path)
data_train_background = get_data_background(logger, data_train_background_batch, project_name=project_name, delete_labels=delete_labels_background, nfs_mount_path=nfs_mount_path)
data_val = get_data_val(logger, data_val_batch, project_name=project_name, delete_labels=delete_labels, nfs_mount_path=nfs_mount_path)
logger.info_ai(meg="download data, train data num:%d, background data num:%d, val data num:%d"%(len(data_train), len(data_train_background), len(data_val)))
#储存每个类别嵌件个数，以便后续裁剪图片时按概率裁剪
class_num_dic = {}
for line in data_train:
pic_path, boxes, labels = parse_line(line)
for label in labels:
if label not in config.data_set["fill_zero_label_names"] and label != "gj":
if label not in class_num_dic:
class_num_dic[label] = 1
else:
class_num_dic[label] += 1
data_train = data_train + data_train_background
for line in data_train:
data_train_save.write(line)
data_train_save.write("\n")
data_train_save.close()
logger.info_ai(meg="train data save in:%s" % data_train_save_path)
if len(data_val) > 0:
for line in data_val:
data_val_save.write(line)
data_val_save.write("\n")
data_val_save.close()
logger.info_ai(meg="val data save in:%s" % data_val_save_path)
return len(data_train), len(data_val), class_num_dic
def parse_gt_rec(gt_filename, target_img_size, letterbox_resize=True):
'''
parse and re-organize the gt info.
return:
gt_dict: dict. Each key is a img_id, the value is the gt bboxes in the corresponding img.
'''
global gt_dict
if not gt_dict:
new_width, new_height = target_img_size
with open(gt_filename, 'r') as f:
for line in f:
img_id, pic_path, boxes, labels, ori_width, ori_height = parse_line(
line)
objects = []
for i in range(len(labels)):
x_min, y_min, x_max, y_max = boxes[i]
label = labels[i]
if letterbox_resize:
resize_ratio = min(new_width / ori_width,
new_height / ori_height)
resize_w = int(resize_ratio * ori_width)
resize_h = int(resize_ratio * ori_height)
dw = int((new_width - resize_w) / 2)
dh = int((new_height - resize_h) / 2)
objects.append([
x_min * resize_ratio + dw,
y_min * resize_ratio + dh,
x_max * resize_ratio + dw,
y_max * resize_ratio + dh, label
])
else:
objects.append([
x_min * new_width / ori_width,
y_min * new_height / ori_height,
x_max * new_width / ori_width,
y_max * new_height / ori_height, label
])
gt_dict[img_id] = objects
return gt_dict
def cut_box_signal_process(data_lines,
process_id,
save_path,
do_extension_boxes,
extension_ratio=1.3,
process_over=None):
globalvar.globalvar.logger.info_ai(meg="start process:%d" % process_id)
save_id = 0
for line in data_lines:
try:
save_id = save_id + 1
pic_path_ori, boxes, labels = parse_line(line)
if len(boxes) == 0:
continue
img = cv2.imread(pic_path_ori)
if img is None:
print("img is not exits:%d" % pic_path_ori)
if do_extension_boxes:
boxes = extension_boxes(img,
boxes,
extension_ratio=extension_ratio)
for i in range(len(boxes)):
save_id = save_id + 1
x0, y0, x1, y1 = boxes[i]
img_save = img[y0:y1, x0:x1, :]
if not os.path.exists(save_path + "/%s" % labels[i]):
os.mkdir(save_path + "/%s" % labels[i])
cut_img_save_path = save_path + "/%s/%s_%d_%d.jpg" % (
labels[i], pic_path_ori.split("/")[-1], process_id,
save_id)
cv2.imwrite(cut_img_save_path, img_save)
globalvar.globalvar.logger.info_ai(
meg="cut box img save to:%s" % cut_img_save_path)
except Exception as ex:
globalvar.globalvar.logger.info_ai(
meg="cut box img find Exception:%s" % ex)
process_over[process_id] = 1
globalvar.globalvar.logger.info_ai(meg="this process id over:%d" %
process_id)
config=config_common,
logger=logger,
project_name="sjht")
else:
logger.info_ai(
meg="data batch is None, use train file and val file")
data_train = open(
config_common.data_set["train_file_path"]).readlines()
train_num = len(data_train)
data_val = open(
config_common.data_set["val_file_path"]).readlines()
val_num = len(data_val)
# 储存每个类别嵌件个数，以便后续裁剪图片时按概率裁剪
class_num_dic = {}
for line in data_train:
pic_path, boxes, labels = parse_line(line)
for label in labels:
if label not in config_common.data_set[
"fill_zero_label_names"] and label != "gj":
if label not in class_num_dic:
class_num_dic[label] = 1
else:
class_num_dic[label] += 1
logger.info_ai(
meg="prepare data over, get all class name from data file",
get_ins={"class_num_dic": class_num_dic})
lr_decay_freq = int(
config_common.model_set["lr_decay_freq_epoch"] * train_num /
(len(gpu_device) * config_common.model_set["batch_size"]))
def generate_crop_txt(self, annotation_txt, txt_dst):
# crop center region
contents = open(annotation_txt, 'r').readlines()
dstfile = open(txt_dst, 'w')
for line in contents:
pic_path, boxes, labels = parse_line(line, is_str=True)
img = get_img(pic_path)
try:
h, w, _ = img.shape
except:
print('wrong', pic_path)
continue
# rimg = np.rot90(img).copy()
img_dstdir, imgname = os.path.split(pic_path)
savepath = os.path.join(img_dstdir, 'crop_' + imgname)
aline = savepath
flag = 0
if h > w:
st = int(h / 2 - w / 2)
rimg = img[st:st + w, :, :]
for i in range(len(labels)):
b = boxes[i]
x1, y1, x2, y2 = b[0], max(1, b[1] - st), b[2], min(
b[3] - st, w - 2)
if (y2 - y1) / (b[3] - b[1]) < 0.2 or (y2 - y1) / w < 0.05:
continue
aline = aline + ' {} {} {} {} {}'.format(
labels[i], x1, y1, x2, y2)
flag = 1
# plot_one_box(img, [b[0], b[1], b[2], b[3]], label='test')
# cv2.imshow('ori', img)
# plot_one_box(rimg, [x1, y1, x2, y2], label='test')
# cv2.imshow('test', rimg)
# cv2.waitKey(0)
# aline = aline + ' ' + str(labels[i]) + ' ' + str(b[0]) + ' ' + str(b[1]) + ' ' + str(b[2]) + ' ' + str(b[3])
elif h < w:
st = int(w / 2 - h / 2)
rimg = img[:, st:st + h, :]
for i in range(len(labels)):
b = boxes[i]
x1, y1, x2, y2 = max(1, b[0] - st), b[1], min(
b[2] - st, h - 2), b[3]
if (x2 - x1) / (b[2] - b[0]) < 0.2 or (x2 - x1) / h < 0.05:
continue
aline = aline + ' {} {} {} {} {}'.format(
labels[i], x1, y1, x2, y2)
flag = 1
# plot_one_box(img, [b[0], b[1], b[2], b[3]], label='test')
# cv2.imshow('ori', img)
# plot_one_box(rimg, [x1, y1, x2, y2], label='test')
# cv2.imshow('test', rimg)
# cv2.waitKey(0)
# aline = aline + ' ' + str(labels[i]) + ' ' + str(b[0]) + ' ' + str(b[1]) + ' ' + str(b[2]) + ' ' + str(b[3])
else:
print('no crop', pic_path)
continue
if flag == 0:
continue
dstfile.write(aline)
dstfile.write('\n')
dstfile.close()
self.concat_txts(annotation_txt, txt_dst)
config=config_common,
logger=logger,
project_name="sjht")
else:
logger.info_ai(
meg="data batch is None, use train file and val file")
data_train = open(
config_common.data_set["train_file_path"]).readlines()
train_num = len(data_train)
data_val = open(
config_common.data_set["val_file_path"]).readlines()
val_num = len(data_val)
# 储存每个类别嵌件个数，以便后续裁剪图片时按概率裁剪
class_num_dic = {}
for line in data_val:
pic_path, boxes, labels = parse_line(line)
for label in labels:
if label not in config_common.data_set[
"fill_zero_label_names"] and label != "gj":
if label not in class_num_dic:
class_num_dic[label] = 1
else:
class_num_dic[label] += 1
logger.info_ai(
meg="prepare data over, get all class name from data file",
get_ins={"class_num_dic": class_num_dic})
detection_scores = list(np.arange(0.05, 0.99, 0.05))
class_scores = list(np.arange(0.05, 0.99, 0.05))
if config_common.test_other_info_set["show_wrong_data"]:
