【数据集格式】coco格式转txt格式

Bertha 。 2023-01-16 09:57 2阅读 0赞

# json\_to\_txt #

txt格式  
环境  
代码  
转后路径  
结果

## txt格式 ##

训练tensorflow版的yolov3需要的数据格式是txt的，具体要求如下

> xxx/xxx.jpg 18.19,6.32,424.13,421.83,20 323.86,2.65,640.0,421.94,20  
> xxx/xxx.jpg 48,240,195,371,11 8,12,352,498,14  
> image\_path x\_min, y\_min, x\_max, y\_max, class\_id x\_min, y\_min ,…, class\_id  
> make sure that x\_max < width and y\_max < height

针对tensorflow-yolov3，它的class\_id 是从0开始到类别数-1. 而我自己的数据集的category\_id部分是1，2，3，4，8，9，10

所以我还在代码中添加boundingBox所属类别部分 做了如下修改

id = boundingBox_per['category_id']                
    	if(id in [8,9,10]):                    		
    		id = id-4                
    	else:                    
    		id = id-1

然后class.name 部分为1，2，3，4，8，9，10

## 环境 ##

需要的环境是 python3  
需要修改的地方：第10.11.12行

filename = 'instances_val.json'#修改成你的json文件名字
    jsondir = '/home/chenbolin/data2/six/annotations/instances_val.json'#修改成你的json文件路径
    img_root = '/home/chenbolin/data2/six/val/' #修改成你的图片路径f = open(jsondir,encoding='utf-8')

## 代码 ##

我们的数据集是coco格式的，也就是要把json文件转成txt格式的，代码如下：

# coding=utf-8import json
    import numpy as np
    from xml.etree.ElementTree import Element, SubElement, tostring
    from xml.dom.minidom import parseString
    import os
    # -----------------
    # txt数据格式：path，xmin,ymin,xmax,ymax
    # 每一行表示一个image
    # --------------------
    filename = 'instances_val.json'#修改成你的json文件名字
    jsondir = '/home/chenbolin/data2/six/annotations/instances_val.json'#修改成你的json文件路径
    img_root = '/home/chenbolin/data2/six/val/' #修改成你的图片路径f = open(jsondir,encoding='utf-8')
    res = f.read()
    data = json.loads(res)
    
    # 保存数据的文件夹
    folder = filename.split('.')[0]+'_txt'
    if not os.path.exists(folder):
        os.mkdir(folder)
    
    
    # 首先得到数据的categories的关键字
    category = data['categories']
    category_id ={ }
    for category_per in category:
        id = category_per['id']
        cls = category_per['name']
        category_id[id] = cls
    
    
    print(category_id)
    # 开始遍历字典，对每一个图像生成xml文件
    imageID_all =[]
    imageID_all_info = { }
    for images_attr in list(data.keys()):
        if  images_attr == 'images':
            # 遍历每一个图像
            for data_per in data[images_attr]:                        
                # 获取图像名字
                image_name = data_per['file_name']
                # 获取图像路径
                #image_route = data_per['coco_url'] 
                image_route = img_root+image_name                         
                # 获取图像的像素和ID
                image_width = data_per['width']
                image_height = data_per['height']
                image_id = data_per['id']
                imageID_all.append(image_id)
                imageID_all_info[image_id]={ 'width':image_width,'height':image_height,'path':image_route,'filename':image_name}
    
    
        elif images_attr == 'annotations':
            # 根据id遍历每张图像的bounding box
            for imageID_per in imageID_all:
                print(imageID_per)
                # 根据图像ID，构建图像基本信息子目录
                # 图像路径
                image_path = imageID_all_info[imageID_per]['path']
                # 每一张图片信息写在txt文件
                # filename1 = imageID_all_info[imageID_per]['filename'].split('.')[0]
                file_write = folder + '/' + filename.split('.')[0] + '.txt'
                # 图像包含了多少个bounding box
                boundingBox_image = [j for j in data[images_attr] if j['image_id']==imageID_per]
                boundingBox_cord =''
                path_cord =''
                if len(boundingBox_image)==0:
                    path_cord = image_path +'/n'
                    with open(file_write, 'a+') as f:
                        f.write(path_cord)
                    continue
                # 输出每张boundging box的坐标信息，以及所属类信息
                for boundingBox_per in boundingBox_image:
                    # 添加boundingBox所属类的id
                    id = boundingBox_per['category_id']
                    # 位置信息转换，x,y,w,h转为xmin,ymin,xmax,ymax
                    x = boundingBox_per['bbox'][0]
                    y = boundingBox_per['bbox'][1]
                    w = boundingBox_per['bbox'][2]
                    h = boundingBox_per['bbox'][3]
                    xmin = str(x+1)
                    ymin = str(y+1)
                    xmax= str(round(x+w,2))
                    ymax=str(round(y+h,2))
                    boundingBox_cord += xmin +','+ymin+','+xmax+','+ymax+','+str(id)+' '
    
    
                boundingBox_cord = boundingBox_cord.rstrip()
                boundingBox_cord += '\n'
                path_cord = image_path + ' '+boundingBox_cord
                with open(file_write, 'a+') as f:
                    f.write(path_cord)

## 转后路径 ##

*  !\[class.names所在路径：  
    /home/chenbolin/tensorflow-yolov3/data/classes/six.names
 *  txt文件所在路径：  
    /home/chenbolin/tensorflow-yolov3/data/dataset/six\_train.txt  
    /home/chenbolin/tensorflow-yolov3/data/dataset/six\_test.txt
 *  图片所在路径： /home/chenbolin/data2/six/train  
    /home/chenbolin/data2/six/test\](https://img-blog.csdnimg.cn/20210425134459175.png?x-oss-process=image/watermark,type\_ZmFuZ3poZW5naGVpdGk,shadow\_10,text\_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xvY2tob3U=,size\_16,color\_FFFFFF,t\_70)

## 结果 ##

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xvY2tob3U_size_16_color_FFFFFF_t_70]  
[转载][Link 1]

[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xvY2tob3U_size_16_color_FFFFFF_t_70]: /images/20221022/42d3f3ae9120484293f987a1c62044f7.png
[Link 1]: https://blog.csdn.net/zxfhahaha/article/details/108024608