自建数据集系列：从RGB-＞二值mask-＞coco格式

前言

前文说到，mask可以转labelme，然后再转coco。但对于实例级的mask事情变得有点不同，需先将RGB的mask拆解为二值mask然后进行直接操作，省略labelme这一步骤。

image文件夹下

annotations文件夹下
模型加载处理好的数据效果

RGB转二值mask

rgb2mask.py

import cv2
import numpy as np
import os, glob
 
part = "test"

def rgb2masks(label_name):
# Camouflaged:
# COD10K-CAM-SuperNumber-SuperClass-SubNumber-SubClass-ImageNumber

# Non-Camouflaged:
# COD10K-NonCAM-SuperNumber-SuperClass-SubNumber-SubClass-ImageNumber

    name = os.path.split(label_name)[-1].split('.')[0]
    lbl_id = name.split('-')[-1]
    subClass = name.split('-')[-2]
    lbl = cv2.imread(label_name, 1)
    try:
        h, w = lbl.shape[:2]
    
        leaf_dict = {}
        idx = 0
        white_mask = np.ones((h, w, 3), dtype=np.uint8) * 255
        for i in range(h):
            for j in range(w):
                if tuple(lbl[i][j]) in leaf_dict or tuple(lbl[i][j]) == (0, 0, 0):
                    continue
                leaf_dict[tuple(lbl[i][j])] = idx
                mask = (lbl == lbl[i][j]).all(-1)
                # leaf = lbl * mask[..., None]      # colorful leaf with black background
                # np.repeat(mask[...,None],3,axis=2)    # 3D mask
                leaf = np.where(mask[..., None], white_mask, 0)
                mask_name = './'+part+'/annotations/' + lbl_id +'_'+subClass +'_'+ str(idx) + '.png'  # ImageNumber_SubClass_idx.png
                cv2.imwrite(mask_name, leaf)
                idx += 1
        print("正常："+label_name)
    except:
        print("cannot read:"+label_name)
        with open(part+"_error.txt",'a+') as f:
            f.write(label_name)
            f.write('\n')
 
label_dir = './instance_'+part
label_list = glob.glob(os.path.join(label_dir, '*.png'))
for label_name in label_list:
    rgb2masks(label_name)

二值mask转coco格式

这块的代码作用有二，1.将原始图片移动并重命名 2.生成coco的json文件

mask2coco.py

import datetime
import json
import os
import re
import fnmatch
from PIL import Image
import numpy as np
from pycococreatortools import pycococreatortools
from glob import glob
import cv2
import shutil
 
part = 'test'
IMAGE_SRC = 'C:/Users/awei/Desktop/rgb2mask/Image_'+part+'/'
ROOT_DIR = 'C:/Users/awei/Desktop/rgb2mask/modify_'+part
IMAGE_DIR = os.path.join(ROOT_DIR, "image")
ANNOTATION_DIR = os.path.join(ROOT_DIR, "annotations")
 
INFO = {
    "description": "Leaf Dataset",
    "url": "https://github.com/waspinator/pycococreator",
    "version": "0.1.0",
    "year": 2017,
    "contributor": "Francis_Liu",
    "date_created": datetime.datetime.utcnow().isoformat(' ')
}
 
LICENSES = [
    {
        "id": 1,
        "name": "Attribution-NonCommercial-ShareAlike License",
        "url": "http://creativecommons.org/licenses/by-nc-sa/2.0/"
    }
]

# 根据自己的需要添加种类
CATEGORIES = [
    # {
    #     'id': 1,  # 是数字1，不是字符串
    #     'name': 'leaf',
    #     'supercategory': 'leaf',
    # }
]

# Camouflaged:
# COD10K-CAM-SuperNumber-SuperClass-SubNumber-SubClass-ImageNumber

# Non-Camouflaged:
# COD10K-NonCAM-SuperNumber-SuperClass-SubNumber-SubClass-ImageNumber
# Super_Class_Dictionary = {'1':'Aquatic', '2':'Terrestrial', '3':'Flying', '4':'Amphibian', '5':'Other'}
# Sub_Class_Dictionary = {'1':'batFish','2':'clownFish','3':'crab','4':'crocodile','5':'crocodileFish','6':'fish','7':'flounder',
#              '8':'frogFish','9':'ghostPipefish','10':'leafySeaDragon','11':'octopus','12':'pagurian','13':'pipefish',
#               '14':'scorpionFish','15':'seaHorse','16':'shrimp','17':'slug','18':'starFish','19':'stingaree',
#               '20':'turtle','21':'ant','22':'bug','23':'cat','24':'caterpillar','25':'centipede','26':'chameleon',
#               '27':'cheetah','28':'deer','29':'dog','30':'duck','31':'gecko','32':'giraffe','33':'grouse','34':'human',
#               '35':'kangaroo','36':'leopard','37':'lion','38':'lizard','39':'monkey','40':'rabbit','41':'reccoon',
#               '42':'sciuridae','43':'sheep','44':'snake','45':'spider','46':'stickInsect','47':'tiger','48':'wolf',
#               '49':'worm','50':'bat','51':'bee','52':'beetle','53':'bird','54':'bittern','55':'butterfly','56':'cicada',
#               '57':'dragonfly','58':'frogmouth','59':'grasshopper','60':'heron','61':'katydid','62':'mantis',
#               '63':'mockingbird','64':'moth','65':'owl','66':'owlfly','67':'frog','68':'toad','69':'other'}

def getCategories():
    image_files = glob(IMAGE_SRC + "*.jpg")
    subClassList = []
    temp = []
    for image in image_files:
        
        image_name = os.path.basename(image).split('.')[0]
        try:
            _,type,superNumer,superClass,subNumber,subClass,imageNumber = image_name.split('-')
        except:
            print("NonCAM")
            continue


        if not type=="CAM":
            continue


        if not os.path.exists(IMAGE_DIR+"/"+str(imageNumber)+".jpg"):
            shutil.copy(image, IMAGE_DIR+"/"+str(imageNumber)+".jpg")
        if subClass not in subClassList:
            subClassList.append(subClass)
            item = {'id':int(subNumber),  # 强转int类型，很重要！！
                    'name':subClass,
                    'supercategory':superClass
            }
            temp.append(item)
    global CATEGORIES
    CATEGORIES = sorted(temp,key=lambda x: x["id"])

    
 
def filter_for_jpeg(root, files):
    file_types = ['*.jpeg', '*.jpg', '*.png']
    file_types = r'|'.join([fnmatch.translate(x) for x in file_types])
    files = [os.path.join(root, f) for f in files]
    files = [f for f in files if re.match(file_types, f)]
    return files
 
 
def filter_for_annotations(root, files, image_filename):
    file_types = ['*.png']
    file_types = r'|'.join([fnmatch.translate(x) for x in file_types])
    basename_no_extension = os.path.splitext(os.path.basename(image_filename))[0]
    file_name_prefix = basename_no_extension + '_.*'   # 用于匹配对应的二值mask
    files = [os.path.join(root, f) for f in files]
    files = [f for f in files if re.match(file_types, f)]
    files = [f for f in files if re.match(file_name_prefix, os.path.splitext(os.path.basename(f))[0])]
    return files
 
 
def main():
    getCategories()
    coco_output = {
        "info": INFO,
        "licenses": LICENSES,
        "categories": CATEGORIES,
        "images": [],
        "annotations": []
    }


 
    image_id = 1
    segmentation_id = 1
 
    # filter for jpeg images
    for root, _, files in os.walk(IMAGE_DIR):
        image_files = filter_for_jpeg(root, files)
 
        # go through each image
        for image_filename in image_files:
            image = Image.open(image_filename)
            image_info = pycococreatortools.create_image_info(
                    image_id, os.path.basename(image_filename), image.size)
            coco_output["images"].append(image_info)
 
            # filter for associated png annotations
            for root, _, files in os.walk(ANNOTATION_DIR):
                annotation_files = filter_for_annotations(root, files, image_filename)
 
                # go through each associated annotation
                for annotation_filename in annotation_files:
 
                    
                    # class_id = [x['id'] for x in CATEGORIES if x['name'] in annotation_filename][0]
                    class_id = [x['id'] for x in CATEGORIES if x['name'].upper() == annotation_filename.split('_')[-2].upper()][0]  # 精确匹配类型名

                    print(annotation_filename+" "+str(class_id))
 
                    category_info = {'id': class_id, 'is_crowd': 'crowd' in image_filename}
                    binary_mask = np.asarray(Image.open(annotation_filename)
                                             .convert('1')).astype(np.uint8)
 
                    annotation_info = pycococreatortools.create_annotation_info(
                            segmentation_id, image_id, category_info, binary_mask,
                            image.size, tolerance=2)
 
                    if annotation_info is not None:
                        coco_output["annotations"].append(annotation_info)
 
                    segmentation_id = segmentation_id + 1
 
            image_id = image_id + 1
 
    with open(ROOT_DIR+'/instances_'+part+'2017.json', 'w') as output_json_file:
        json.dump(coco_output, output_json_file)
 
 
if __name__ == "__main__":
    main()

汇总

1.从labelImg格式->txt格式（YOLO格式、ICDAR2015格式）

2.从二值mask-＞labelme格式-＞coco格式

3.从labelme格式-＞VOC格式+从二值mask-＞VOC格式

4.从RGB-＞二值mask-＞coco格式

5.实例分割mask-＞语义分割mask-＞扩增mask

6.COCO格式-＞YOLO格式

双模图片数据与对应标注文件的命名对齐

xml标注文件的节点、属性、文本的修正

cocoJson数据集统计分析