Create own dataset

[bharatkumarreddy]

Hi,

Is there any tool to recreate the XML in Pascal VOC format. I have found few on github but they are for Rectangles. Can u suggest any ?

[MhLiao]

@bharatkumarreddy You can use "xml.dom.minidom" in Python to create xml files. Here is an example:

import xml.dom.minidom as Dom 
def create_xml_rotated_icdar13_train(root_dir, save_dir, generate_file):
    generate_list=open(generate_file,'wt')
    count=0
    # for i in range(1000):
    for root, dirs, files in os.walk(os.path.join(root_dir,'train_images/')):
        for file in files:
            image_name=file
            label_name='gt_'+file.split('.')[0]+'.txt'
            image_path=os.path.join(root_dir,'train_images/',image_name)
            sub_dir='train_images'
            print(image_path)
            I=Image.open(image_path)
            image_width,image_height=I.size
            label_path=os.path.join(root_dir,'train_gts/',label_name)
            label=open(label_path,'r')
            doc = Dom.Document() 
            root_node=doc.createElement('annotation')
            doc.appendChild(root_node)
            folder_node=doc.createElement('folder')
            folder_node_value=doc.createTextNode(sub_dir)
            folder_node.appendChild(folder_node_value)
            filename_node=doc.createElement('filename')
            filename_value=doc.createTextNode(image_name)
            filename_node.appendChild(filename_value)
            size_node=doc.createElement('size')
            root_node.appendChild(size_node)
            width_node=doc.createElement('width')
            width_value=doc.createTextNode(str(image_width))
            width_node.appendChild(width_value)
            height_node=doc.createElement('height')
            height_value=doc.createTextNode(str(image_height))
            height_node.appendChild(height_value)
            depth_node=doc.createElement('depth')
            depth_value=doc.createTextNode('3')
            depth_node.appendChild(depth_value)
            size_node.appendChild(width_node)
            size_node.appendChild(height_node)
            size_node.appendChild(depth_node)
            first_line=True
            count=0
            for line in label.readlines():
                if first_line:
                    line=line.decode('utf-8-sig').strip()
                else:
                    line=line.strip()
                print(line)
                bbs=line.split(' ')
                xmin=int(bbs[0])
                ymin=int(bbs[1])
                xmax=int(bbs[2])
                ymax=int(bbs[3])
                x1=xmin
                x2=xmax
                x3=xmax
                x4=xmin
                y1=ymin
                y2=ymin
                y3=ymax
                y4=ymax
                xmin=str(min(x1,x2,x3,x4))
                ymin=str(min(y1,y2,y3,y4))
                xmax=str(max(x1,x2,x3,x4))
                ymax=str(max(y1,y2,y3,y4))
                x1=str(x1)
                x2=str(x2)
                x3=str(x3)
                x4=str(x4)
                y1=str(y1)
                y2=str(y2)
                y3=str(y3)
                y4=str(y4)
                content=bbs[4]
                content=''.join(ch for ch in content if ch.isalnum())
                content=content.upper()
                if content=='':
                    content='###'
                    difficult='1'
                else:
                    difficult='0'
                object_node=doc.createElement('object')
                root_node.appendChild(object_node)
                name_node=doc.createElement('name')
                name_value=doc.createTextNode('text')
                name_node.appendChild(name_value)
                content_node=doc.createElement('content')
                content_value=doc.createTextNode(content)
                content_node.appendChild(content_value)
                difficult_node=doc.createElement('difficult')
                difficult_value=doc.createTextNode(difficult)
                difficult_node.appendChild(difficult_value)
                bndbox_node=doc.createElement('bndbox')
                object_node.appendChild(bndbox_node)
                x1_node=doc.createElement('x1')
                x1_value=doc.createTextNode(x1)
                x1_node.appendChild(x1_value)
                y1_node=doc.createElement('y1')
                y1_value=doc.createTextNode(y1)
                y1_node.appendChild(y1_value)
                x2_node=doc.createElement('x2')
                x2_value=doc.createTextNode(x2)
                x2_node.appendChild(x2_value)
                y2_node=doc.createElement('y2')
                y2_value=doc.createTextNode(y2)
                y2_node.appendChild(y2_value)
                x3_node=doc.createElement('x3')
                x3_value=doc.createTextNode(x3)
                x3_node.appendChild(x3_value)
                y3_node=doc.createElement('y3')
                y3_value=doc.createTextNode(y3)
                y3_node.appendChild(y3_value)
                x4_node=doc.createElement('x4')
                x4_value=doc.createTextNode(x4)
                x4_node.appendChild(x4_value)
                y4_node=doc.createElement('y4')
                y4_value=doc.createTextNode(y4)
                y4_node.appendChild(y4_value)
                xmin_node=doc.createElement('xmin')
                xmin_value=doc.createTextNode(xmin)
                xmin_node.appendChild(xmin_value)
                ymin_node=doc.createElement('ymin')
                ymin_value=doc.createTextNode(ymin)
                ymin_node.appendChild(ymin_value)
                xmax_node=doc.createElement('xmax')
                xmax_value=doc.createTextNode(xmax)
                xmax_node.appendChild(xmax_value)
                ymax_node=doc.createElement('ymax')
                ymax_value=doc.createTextNode(ymax)
                ymax_node.appendChild(ymax_value)
                bndbox_node.appendChild(x1_node)
                bndbox_node.appendChild(y1_node)
                bndbox_node.appendChild(x2_node)
                bndbox_node.appendChild(y2_node)
                bndbox_node.appendChild(x3_node)
                bndbox_node.appendChild(y3_node)
                bndbox_node.appendChild(x4_node)
                bndbox_node.appendChild(y4_node)
                bndbox_node.appendChild(xmin_node)
                bndbox_node.appendChild(ymin_node)
                bndbox_node.appendChild(xmax_node)
                bndbox_node.appendChild(ymax_node)
                object_node.appendChild(difficult_node)
                object_node.appendChild(content_node)
                object_node.appendChild(name_node)
                object_node.appendChild(bndbox_node)
                count=count+1
            root_node.appendChild(folder_node)
            root_node.appendChild(filename_node)
            root_node.appendChild(size_node)
            for i in range(0,count): 
                root_node.appendChild(object_node)
            if count>0:
                xml_file_path=save_dir+image_name[0:len(image_name)-4]+'.xml'
                f = open(xml_file_path, 'w+') 
                f.write(doc.toprettyxml(indent = "\t", newl = "\n", encoding = "utf-8")) 
                f.close()
                generate_list.write(image_path+' '+xml_file_path+'\n')

            label.close()
    generate_list.close()

[19931991]

Do you have recreate the new XML? I hope that there is a chance to ask you , and my qq is 1323369151, looking forward for your adding my qq ! Thank you very much!@ bharatkumarreddy

[19931991]

Ground truth can be directly converted to xml by the code ? @MhLiao

[MhLiao]

@19931991 Yes, it is suitable for the ground truth of ICDAR 2013. For other formats such as ICDAR 2015, you should modify the code to get the corresponding coordinates.

[19931991]

Thanks for your reply ! I change the input path to recreate the xml . but it creates nothing , the file is null , this is my code of xml , I want to know where is the problem ?@MhLiao

[19931991]

I have solved the problem, the net is running , thank you very much!@MhLiao

[khalilo]

I also find nothing as a result. @19931991 how did you solve the problem ?