08 semi automatic annotation of target detection data set

Deep learning target detection usually requires a large data scale , But the annotation of data sets is usually a time-consuming and laborious thing with little significance , We can mark three or five hundred pictures first , Then train a relatively preliminary model , Then use this model to infer the unlabeled image , Export the reasoning result as voc Format , Then use the local image annotation software to annotate the pictures and annotation files labelimg open , After manually adjusting the position of the border , The annotation file can be used as a data set for subsequent model training .

Reasoning

!python /home/aistudio/PaddleDetection/tools/infer.py \
	-c /home/aistudio/PaddleDetection/configs/cascade_rcnn/cascade_rcnn_r50_fpn_1x_coco.yml \
	--draw_threshold=0.5 \
	--infer_dir=img \
	--output_dir=toolinfer \
	--use_vdl=True \
	--save_txt=True \
	-o weights=/home/aistudio/output/cascade_rcnn_r50_fpn_1x_coco/model_final

Save the reasoning results to txt In file

The reasoning result is modified as voc Format annotation

Move the generated inference result file to the specified folder , Modify the corresponding coordinate information , Generate voc Format xml Mark the file .

Convert code

import os
import cv2

headstr = """\ <annotation> <folder>VOC</folder> <filename>%s</filename> <source> <database>My Database</database> <annotation>VOC</annotation> <image>flickr</image> <flickrid>NULL</flickrid> </source> <owner> <flickrid>NULL</flickrid> <name>company</name> </owner> <size> <width>%d</width> <height>%d</height> <depth>%d</depth> </size> <segmented>0</segmented> """
objstr = """\ <object> <name>%s</name> <pose>Unspecified</pose> <truncated>0</truncated> <difficult>%d</difficult> <bndbox> <xmin>%d</xmin> <ymin>%d</ymin> <xmax>%d</xmax> <ymax>%d</ymax> </bndbox> </object> """
tailstr = '''\ </annotation> '''


def save_annotations(boxes, img, filename):
    H = img.shape[0]
    W = img.shape[1]
    C = img.shape[2]
    # H,W,C = img.shape
    img_name = filename.split('.')[0] + '.bmp'
    head = headstr % (img_name, W, H, C)  #  Write header file 
    tail = tailstr  #  Write tail file 
    #  write in boxes
    save_path = anno_path + filename.split('.')[0] + '.xml'
    f = open(save_path, 'w')
    f.write(head)
    for box in boxes:
        f.write(objstr % (str(box[0]), 0, float(box[2]), float(box[3]), float(box[2]) + float(box[4]), float(box[3]) + float(box[5])))
    f.write(tail)


if __name__ == '__main__':
    #  Set the path 
    root_path = './'
    total_label_path = root_path + 'txt/'  # txt Storage path 
    total_img_path = root_path + 'img/'  #  Image storage path 
    anno_path = root_path + 'Annotations/'  #  Store generated xml Mark the file 
    #  Judge whether there is... In the current path Annotations This folder , If it does not exist , Automatically create a 
    if not os.path.exists(anno_path):
        os.mkdir(anno_path)
    #  Read one by one txt Mark the file 
    for filename in os.listdir(total_label_path):
        cur_label_path = total_label_path + filename
        cur_img_path = total_img_path + filename.split('.')[0] + '.bmp'  #  Change the file name suffix 
        cur_boxes = []
        #  Read the current txt Contents of the file 
        with open(cur_label_path, 'r') as file:
            while True:
                line = file.readline().strip()  # .strip() To remove '\r,\n'
                if not line:
                    break
                line_list = [ele for ele in line.split(' ')]
                cur_boxes.append(line_list)
        #  Read the current image 
        cur_img = cv2.imread(cur_img_path)
        #  Conduct xml Document storage 
        save_annotations(cur_boxes, cur_img, filename)

What needs to be revised ：

1. txt Path of file storage total_label_path
2. Image storage path total_img_path
3. Create and store voc Format folder anno_path
4. The suffix of the image file bmp

matters needing attention ：

1. Generally, three or five hundred images with initial annotation are required
2. Try to use high-precision target detection algorithm to complete pre training
3. According to the prediction results and the situation of the real target , Reasonably select the reasoning threshold . If there are many missed inspections , Then reduce the confidence threshold ; If there are many false inspections , Then increase the confidence threshold .