diff --git a/README.md b/README.md index 7abe882..a108f50 100644 --- a/README.md +++ b/README.md @@ -4,7 +4,7 @@ - [ ] 数据集转换为统一格式(检测和识别) - [x] icdar2015 - [x] MLT2019 - - [ ] COCO-Text_v2 + - [x] COCO-Text_v2 - [ ] ReCTS - [ ] SROIE - [ ] ArT diff --git a/convert/check_json.py b/convert/check_json.py index 0c1bc65..3f70ecc 100644 --- a/convert/check_json.py +++ b/convert/check_json.py @@ -1,6 +1,9 @@ # -*- coding: utf-8 -*- # @Time : 2020/3/20 20:33 # @Author : zhoujun +""" +用于检查生成的json文件有没有问题 +""" from PIL import Image from matplotlib import pyplot as plt diff --git a/convert/coco_text.py b/convert/coco_text.py new file mode 100644 index 0000000..2e8182b --- /dev/null +++ b/convert/coco_text.py @@ -0,0 +1,237 @@ +from __future__ import absolute_import +from __future__ import division +from __future__ import print_function + +__author__ = 'andreasveit' +__version__ = '2.0' +# Interface for accessing the COCO-Text dataset. + +# COCO-Text is a large dataset designed for text detection and recognition. +# This is a Python API that assists in loading, parsing and visualizing the +# annotations. The format of the COCO-Text annotations is also described on +# the project website http://vision.cornell.edu/se3/coco-text/. In addition to this API, please download both +# the COCO images and annotations. +# This dataset is based on Microsoft COCO. Please visit http://mscoco.org/ +# for more information on COCO, including for the image data, object annotatins +# and caption annotations. + +# An alternative to using the API is to load the annotations directly +# into Python dictionary: +# with open(annotation_filename) as json_file: +# coco_text = json.load(json_file) +# Using the API provides additional utility functions. + +# The following API functions are defined: +# COCO_Text - COCO-Text api class that loads COCO annotations and prepare data structures. +# getAnnIds - Get ann ids that satisfy given filter conditions. +# getImgIds - Get img ids that satisfy given filter conditions. +# loadAnns - Load anns with the specified ids. +# loadImgs - Load imgs with the specified ids. +# loadRes - Load algorithm results and create API for accessing them. +# Throughout the API "ann"=annotation, "cat"=category, and "img"=image. + +# COCO-Text Toolbox. Version 1.1 +# Data and paper available at: http://vision.cornell.edu/se3/coco-text/ +# Code based on Microsoft COCO Toolbox Version 1.0 by Piotr Dollar and Tsung-Yi Lin +# extended and adapted by Andreas Veit, 2016. +# Licensed under the Simplified BSD License [see bsd.txt] + +import json +import datetime +import matplotlib.pyplot as plt +from matplotlib.collections import PatchCollection +from matplotlib.patches import Rectangle, PathPatch +from matplotlib.path import Path +import numpy as np +import copy +import os + +class COCO_Text: + def __init__(self, annotation_file=None): + """ + Constructor of COCO-Text helper class for reading and visualizing annotations. + :param annotation_file (str): location of annotation file + :return: + """ + # load dataset + self.dataset = {} + self.anns = {} + self.imgToAnns = {} + self.catToImgs = {} + self.imgs = {} + self.cats = {} + self.val = [] + self.test = [] + self.train = [] + if not annotation_file == None: + assert os.path.isfile(annotation_file), "file does not exist" + print('loading annotations into memory...') + time_t = datetime.datetime.utcnow() + dataset = json.load(open(annotation_file, 'r')) + print(datetime.datetime.utcnow() - time_t) + self.dataset = dataset + self.createIndex() + + def createIndex(self): + # create index + print('creating index...') + self.imgToAnns = {int(cocoid): self.dataset['imgToAnns'][cocoid] for cocoid in self.dataset['imgToAnns']} + self.imgs = {int(cocoid): self.dataset['imgs'][cocoid] for cocoid in self.dataset['imgs']} + self.anns = {int(annid): self.dataset['anns'][annid] for annid in self.dataset['anns']} + self.cats = self.dataset['cats'] + self.val = [int(cocoid) for cocoid in self.dataset['imgs'] if self.dataset['imgs'][cocoid]['set'] == 'val'] + self.test = [int(cocoid) for cocoid in self.dataset['imgs'] if self.dataset['imgs'][cocoid]['set'] == 'test'] + self.train = [int(cocoid) for cocoid in self.dataset['imgs'] if self.dataset['imgs'][cocoid]['set'] == 'train'] + print('index created!') + + def info(self): + """ + Print information about the annotation file. + :return: + """ + for key, value in self.dataset['info'].items(): + print('%s: %s'%(key, value)) + + def filtering(self, filterDict, criteria): + return [key for key in filterDict if all(criterion(filterDict[key]) for criterion in criteria)] + + def getAnnByCat(self, properties): + """ + Get ann ids that satisfy given properties + :param properties (list of tuples of the form [(category type, category)] e.g., [('readability','readable')] + : get anns for given categories - anns have to satisfy all given property tuples + :return: ids (int array) : integer array of ann ids + """ + return self.filtering(self.anns, [lambda d, x=a, y=b:d[x] == y for (a,b) in properties]) + + def getAnnIds(self, imgIds=[], catIds=[], areaRng=[]): + """ + Get ann ids that satisfy given filter conditions. default skips that filter + :param imgIds (int array) : get anns for given imgs + catIds (list of tuples of the form [(category type, category)] e.g., [('readability','readable')] + : get anns for given cats + areaRng (float array) : get anns for given area range (e.g. [0 inf]) + :return: ids (int array) : integer array of ann ids + """ + imgIds = imgIds if type(imgIds) == list else [imgIds] + catIds = catIds if type(catIds) == list else [catIds] + + if len(imgIds) == len(catIds) == len(areaRng) == 0: + anns = list(self.anns.keys()) + else: + if not len(imgIds) == 0: + anns = sum([self.imgToAnns[imgId] for imgId in imgIds if imgId in self.imgToAnns],[]) + else: + anns = list(self.anns.keys()) + anns = anns if len(catIds) == 0 else list(set(anns).intersection(set(self.getAnnByCat(catIds)))) + anns = anns if len(areaRng) == 0 else [ann for ann in anns if self.anns[ann]['area'] > areaRng[0] and self.anns[ann]['area'] < areaRng[1]] + return anns + + def getImgIds(self, imgIds=[], catIds=[]): + ''' + Get img ids that satisfy given filter conditions. + :param imgIds (int array) : get imgs for given ids + :param catIds (int array) : get imgs with all given cats + :return: ids (int array) : integer array of img ids + ''' + imgIds = imgIds if type(imgIds) == list else [imgIds] + catIds = catIds if type(catIds) == list else [catIds] + + if len(imgIds) == len(catIds) == 0: + ids = list(self.imgs.keys()) + else: + ids = set(imgIds) + if not len(catIds) == 0: + ids = ids.intersection(set([self.anns[annid]['image_id'] for annid in self.getAnnByCat(catIds)])) + return list(ids) + + def loadAnns(self, ids=[]): + """ + Load anns with the specified ids. + :param ids (int array) : integer ids specifying anns + :return: anns (object array) : loaded ann objects + """ + if type(ids) == list: + return [self.anns[id] for id in ids] + elif type(ids) == int: + return [self.anns[ids]] + + def loadImgs(self, ids=[]): + """ + Load anns with the specified ids. + :param ids (int array) : integer ids specifying img + :return: imgs (object array) : loaded img objects + """ + if type(ids) == list: + return [self.imgs[id] for id in ids] + elif type(ids) == int: + return [self.imgs[ids]] + + def showAnns(self, anns, show_mask=False): + """ + Display the specified annotations. + :param anns (array of object): annotations to display + :return: None + """ + if len(anns) == 0: + return 0 + ax = plt.gca() + boxes = [] + color = [] + for ann in anns: + c = np.random.random((1, 3)).tolist()[0] + if show_mask: + verts = list(zip(*[iter(ann['mask'])] * 2)) + [(0, 0)] + codes = [Path.MOVETO] + [Path.LINETO] * (len(verts) - 2) + [Path.CLOSEPOLY] + path = Path(verts, codes) + patch = PathPatch(path, facecolor='none') + boxes.append(patch) + text_x, text_y = verts[0] + else: + left, top, width, height = ann['bbox'] + boxes.append(Rectangle([left,top],width,height,alpha=0.4)) + text_x, text_y = left, top + color.append(c) + if 'utf8_string' in list(ann.keys()): + ax.annotate(ann['utf8_string'],(text_x, text_y-4),color=c) + p = PatchCollection(boxes, facecolors=color, edgecolors=(0,0,0,1), linewidths=3, alpha=0.4) + ax.add_collection(p) + + def loadRes(self, resFile): + """ + Load result file and return a result api object. + :param resFile (str) : file name of result file + :return: res (obj) : result api object + """ + res = COCO_Text() + res.dataset['imgs'] = [img for img in self.dataset['imgs']] + + print('Loading and preparing results... ') + time_t = datetime.datetime.utcnow() + if type(resFile) == str: + anns = json.load(open(resFile)) + else: + anns = resFile + assert type(anns) == list, 'results in not an array of objects' + annsImgIds = [int(ann['image_id']) for ann in anns] + + if set(annsImgIds) != (set(annsImgIds) & set(self.getImgIds())): + print('Results do not correspond to current coco set') + print('skipping ', str(len(set(annsImgIds)) - len(set(annsImgIds) & set(self.getImgIds()))), ' images') + annsImgIds = list(set(annsImgIds) & set(self.getImgIds())) + + res.imgToAnns = {cocoid : [] for cocoid in annsImgIds} + res.imgs = {cocoid: self.imgs[cocoid] for cocoid in annsImgIds} + + assert anns[0]['bbox'] != [], 'results have incorrect format' + for id, ann in enumerate(anns): + if ann['image_id'] not in annsImgIds: + continue + bb = ann['bbox'] + ann['area'] = bb[2]*bb[3] + ann['id'] = id + res.anns[id] = ann + res.imgToAnns[ann['image_id']].append(id) + print('DONE (t=%0.2fs)'%((datetime.datetime.utcnow() - time_t).total_seconds())) + + return res \ No newline at end of file diff --git a/convert/coco_text2json.py b/convert/coco_text2json.py new file mode 100644 index 0000000..20ef666 --- /dev/null +++ b/convert/coco_text2json.py @@ -0,0 +1,72 @@ +# -*- coding: utf-8 -*- +# @Time : 2020/3/21 12:54 +# @Author : zhoujun +""" +将coco_text数据集转换为统一格式 +""" +import os +import numpy as np +from tqdm import tqdm +from convert.utils import load, save +from convert.coco_text import COCO_Text + +def cvt(gt_path, save_path, imgs_folder): + gt_dict = {'data_root': imgs_folder} + data_list = [] + ct = COCO_Text(gt_path) + + train_img_ids = ct.getImgIds(imgIds=ct.val) + for img_id in tqdm(train_img_ids): + img = ct.loadImgs(img_id)[0] + # img_path = os.path.join(imgs_folder, img['file_name']) + # if not os.path.exists(img_path): + # continue + cur_gt = {'img_name': img['file_name'], 'annotations': []} + annIds = ct.getAnnIds(imgIds=img['id']) + anns = ct.loadAnns(annIds) + for ann in anns: + if len(ann['utf8_string']) == 0: + continue + cur_line_gt = {'polygon': [], 'text': '', 'illegibility': False, 'language': 'Latin'} + chars_gt = [{'polygon': [], 'char': '', 'illegibility': False, 'language': 'Latin'}] + cur_line_gt['chars'] = chars_gt + + cur_line_gt['language'] = ann['language'] + chars_gt[0]['language'] = ann['language'] + + cur_line_gt['polygon'] = np.array(ann['mask']).reshape(-1,2).tolist() + cur_line_gt['text'] = ann['utf8_string'] + cur_line_gt['illegibility'] = True if ann['legibility'] == "illegible" else False + cur_gt['annotations'].append(cur_line_gt) + if len(cur_gt['annotations']) > 0: + data_list.append(cur_gt) + gt_dict['data_list'] = data_list + save(gt_dict, save_path) + print(len(gt_dict), len(data_list)) + + +def show_coco(gt_path, imgs_folder): + import numpy as np + import skimage.io as io + import matplotlib.pyplot as plt + + data = COCO_Text(gt_path) + # get all images containing at least one instance of legible text + imgIds = data.getImgIds(imgIds=data.train) + # pick one at random + img = data.loadImgs(imgIds[np.random.randint(0, len(imgIds))])[0] + I = io.imread(os.path.join(imgs_folder, img['file_name'])) + plt.figure() + plt.imshow(I) + annIds = data.getAnnIds(imgIds=img['id']) + anns = data.loadAnns(annIds) + data.showAnns(anns) + plt.show() + + +if __name__ == '__main__': + gt_path = r'D:\dataset\COCO_Text\detection\cocotext.v2.json' + imgs_folder = r'D:\dataset\COCO_Text\detection\val' + save_path = r'D:\dataset\COCO_Text\detection\val.json' + cvt(gt_path, save_path, imgs_folder) + # show_coco(gt_path, imgs_folder) diff --git a/convert/convert2jpg.py b/convert/convert2jpg.py index e37ba5d..fa82198 100644 --- a/convert/convert2jpg.py +++ b/convert/convert2jpg.py @@ -1,6 +1,9 @@ # -*- coding: utf-8 -*- # @Time : 2020/3/21 10:37 # @Author : zhoujun +""" +用于将图片统一转换为jpg +""" import os import pathlib from tqdm import tqdm diff --git a/convert/crop_rec.py b/convert/crop_rec.py index dbe356a..8456371 100644 --- a/convert/crop_rec.py +++ b/convert/crop_rec.py @@ -1,6 +1,9 @@ # -*- coding: utf-8 -*- # @Time : 2020/3/20 20:55 # @Author : zhoujun +""" +根据生成的json文件 裁剪出识别训练数据 +""" import os import cv2 import shutil @@ -63,8 +66,8 @@ def four_point_transform(image, pts): if __name__ == '__main__': - json_path = r'D:\dataset\mlt2019\detection\gt.json' - save_path = r'D:\dataset\mlt2019\recognition\train' + json_path = r'D:\dataset\COCO_Text\detection\train.json' + save_path = r'D:\dataset\COCO_Text\recognition\train' gt_path = pathlib.Path(save_path).parent / 'train.txt' if os.path.exists(save_path): shutil.rmtree(save_path, ignore_errors=True) diff --git a/convert/icdar2015tojson.py b/convert/icdar2015tojson.py index 5fab188..5fc96fc 100644 --- a/convert/icdar2015tojson.py +++ b/convert/icdar2015tojson.py @@ -1,7 +1,9 @@ # -*- coding: utf-8 -*- # @Time : 2020/3/18 14:12 # @Author : zhoujun -import glob +""" +将icdar2015数据集转换为统一格式 +""" import pathlib from tqdm import tqdm from convert.utils import load, save, get_file_list diff --git a/convert/mlt2019tojson.py b/convert/mlt2019tojson.py index 6566ae4..c58c172 100644 --- a/convert/mlt2019tojson.py +++ b/convert/mlt2019tojson.py @@ -1,6 +1,9 @@ # -*- coding: utf-8 -*- # @Time : 2020/3/18 14:12 # @Author : zhoujun +""" +将mlt2019数据集转换为统一格式 +""" import glob import pathlib from tqdm import tqdm diff --git a/convert/move_imgs.py b/convert/move_imgs.py new file mode 100644 index 0000000..9ecceef --- /dev/null +++ b/convert/move_imgs.py @@ -0,0 +1,19 @@ +# -*- coding: utf-8 -*- +# @Time : 2020/3/21 16:17 +# @Author : zhoujun +""" +根据json，将图片移动到指定文件夹，以便删除不需要的图片 +""" +import os +import shutil +from tqdm import tqdm +from convert.utils import load_gt + +if __name__ == '__main__': + json_path = r'D:\dataset\COCO_Text\detection\val.json' + save_path = r'D:\dataset\COCO_Text\detection\val' + os.makedirs(save_path,exist_ok=True) + data = load_gt(json_path) + for img_path, gt in tqdm(data.items()): + dst_path = os.path.join(save_path,os.path.basename(img_path)) + shutil.move(img_path,dst_path) diff --git a/gt_detection.json b/gt_detection.json index 138a334..4baeaef 100644 --- a/gt_detection.json +++ b/gt_detection.json @@ -25,6 +25,7 @@ ], "text": "label", "illegibility":false, + "language":"Latin", "chars": [ { "polygon": [ @@ -46,7 +47,8 @@ ] ], "char": "c", - "illegibility": false + "illegibility": false, + "language":"Latin" } ] }