一、标签转换

我们在使用labeme标签工具，标注完数据后会获得json文件。在标注结束过后，我们需要通过标签转换操作，生成jpg格式原始图片和png格式mask标签图。

1.1 使用img_b64_to_arr将json标签中二进制图像数据变成numpy格式数据，再变成jpg图像

/************ Anaconda3\Lib\site-packages\labelme\utils\image.py *************/
def img_b64_to_arr(img_b64):
    img_data = base64.b64decode(img_b64)
    img_arr = img_data_to_arr(img_data)
    return img_arr

# 将输出结果保存，
PIL.Image.fromarray(img).save(osp.join(out_jpgs_path, file_name.split(".")[0]+'.jpg'))

1.2 使用shapes_to_label将分割标签数据变成单通道png mask标签图

/************ Anaconda3\Lib\site-packages\labelme\utils\shape.py *************/
def shapes_to_label(img_shape, shapes, label_name_to_value):
    cls = np.zeros(img_shape[:2], dtype=np.int32)
    ins = np.zeros_like(cls)
    instances = []
    for shape in shapes:
        points = shape["points"]
        label = shape["label"]
        group_id = shape.get("group_id")
        if group_id is None:
            group_id = uuid.uuid1()
        shape_type = shape.get("shape_type", None)

        cls_name = label
        instance = (cls_name, group_id)

        if instance not in instances:
            instances.append(instance)
        ins_id = instances.index(instance) + 1
        cls_id = label_name_to_value[cls_name]

        mask = shape_to_mask(img_shape[:2], points, shape_type)
        cls[mask] = cls_id
        ins[mask] = ins_id

    return cls, ins

1.3 语义分割标签转换完整代码

def json2mask(json_file,img_file,filter_cls:list,out_jpgs_path,out_mask_path):
    assert json_file.endswith(".json") 
    file_name = Path(json_file).stem 
    
    if os.path.isfile(json_file):
        data = json.load(open(json_file))

        # 获取json里面的图片数据，也就是二进制数据
        imageData = data.get("imageData")
        # 如果通过data.get获取到的数据为空，就重新读取图片数据
        if not imageData:
            # imagePath = os.path.join(json_file, data["imagePath"])
            imagePath = img_file
            with open(imagePath, "rb") as f:
                imageData = f.read()
                imageData = base64.b64encode(imageData).decode("utf-8")
        #  将二进制数据转变成numpy格式的数据
        img = utils.img_b64_to_arr(imageData)

        if len(filter_cls):
            data["shapes"]=[data_item for data_item in data["shapes"] if not data_item['label'] in filter_cls]

        label_name_2_id =label_name_2_id_sample
        #  将标签数据变成单通道的png mask图
        lbl, _ = utils.shapes_to_label(img.shape, data["shapes"], label_name_2_id)

        label_names = [None] * (max(label_name_2_id.values()) + 1)
        for name, value in label_name_2_id.items():
            label_names[value] = name

        
        lbl_viz = imgviz.label2rgb(
            label=lbl, image=imgviz.asgray(img), label_names=label_names, loc="rb"
        )

        # 将输出结果保存，
        PIL.Image.fromarray(img).save(osp.join(out_jpgs_path, file_name.split(".")[0]+'.jpg'))
        utils.lblsave(osp.join(out_mask_path, "%s.png" % file_name.split(".")[0]), lbl)

二、验证、训练数据集划分

在经过标签转换后，生成原始jpg图片和png mask标签图。我们还需要进一步进行验证集，训练集划分。

def split_mask(mask_path_list,Base_mask,trainval_percent,train_percent):
    # random.sample随机划分
    num     = len(mask_path_list) 
    _indexes    = range(num)  
    tv      = int(num*trainval_percent)  
    tr      = int(tv*train_percent)  
    trainval= random.sample(_indexes,tv)  
    train   = random.sample(trainval,tr)  
    
    print("train and val size",tv)
    print("train size",tr)
    
    # 创建txt文件
    ftrainval   = open(os.path.join(Base_mask,'trainval.txt'), 'w')  
    ftest       = open(os.path.join(Base_mask,'test.txt'), 'w')  
    ftrain      = open(os.path.join(Base_mask,'train.txt'), 'w')  
    fval        = open(os.path.join(Base_mask,'val.txt'), 'w')  
    
    # 写入txt文件
    for i in _indexes:  
        name = mask_path_list[i].split('.')[0]+'\n'  
        if i in trainval:  
            ftrainval.write(name)  
            if i in train:  
                ftrain.write(name)  
            else:  
                fval.write(name)  
        else:  
            ftest.write(name)  
    # 关闭txt文件
    ftrainval.close()  
    ftrain.close()  
    fval.close()   

生成的验证集，训练集路径txt文件