将widerface标注转换为VOC格式

将widerface标注转换为VOC格式

原文：https://blog.csdn.net/minstyrain/article/details/77986262

widerface是包含了3万多张总计近40万张人脸的人脸检测库，里面包含了大大小小各式各样的人脸，是不可多得的素材。

请将下面的代码保存至widerface.py，并至于下图所示的eval_tools文件夹下，其他的文件结构一并如图所示。

Update:

由于widerface里包含很多小脸，用SSD训练不一定能收敛，此外SSD要求输入为方形，不然会挤压图片造成变形，因此需要对此做些处理.

1. import os,h5py,cv2,sys,shutil

2. import numpy as np

3. from xml.dom.minidom import Document

4. rootdir="../"

5. convet2yoloformat=True

6. convert2vocformat=True

7. resized_dim=(48, 48)

8.  

9. #最小取20大小的脸，并且补齐

10. minsize2select=20

11. usepadding=True

12.  

13. datasetprefix="/home/yanhe/data/widerface"#

14. def gen_hdf5():

15. imgdir=rootdir+"/WIDER_train/images"

16. gtfilepath=rootdir+"/wider_face_split/wider_face_train_bbx_gt.txt"

17. index =0

18. with open(gtfilepath,'r') as gtfile:

19. faces=[]

20. labels=[]

21. while(True ):#and len(faces)<10

22. imgpath=gtfile.readline()[:-1]

23. if(imgpath==""):

24. break;

25. print index,imgpath

26. img=cv2.imread(imgdir+"/"+imgpath)

27. numbbox=int(gtfile.readline())

28. bbox=[]

29. for i in range(numbbox):

30. line=gtfile.readline()

31. line=line.split()

32. line=line[0:4]

33. if(int(line[3])<=0 or int(line[2])<=0):

34. continue

35. bbox=(int(line[0]),int(line[1]),int(line[2]),int(line[3]))

36. face=img[int(line[1]):int(line[1])+int(line[3]),int(line[0]):int(line[0])+int(line[2])]

37. face=cv2.resize(face, resized_dim)

38. faces.append(face)

39. labels.append(1)

40. cv2.rectangle(img,(int(line[0]),int(line[1])),(int(line[0])+int(line[2]),int(line[1])+int(line[3])),(255,0,0))

41. #cv2.imshow("img",img)

42. #cv2.waitKey(1)

43. index=index+1

44. faces=np.asarray(faces)

45. labels=np.asarray(labels)

46. f=h5py.File('train.h5','w')

47. f['data']=faces.astype(np.float32)

48. f['label']=labels.astype(np.float32)

49. f.close()

50. def viewginhdf5():

51. f = h5py.File('train.h5','r')

52. f.keys()

53. faces=f['data'][:]

54. for face in faces:

55. face=face.astype(np.uint8)

56. cv2.imshow("img",face)

57. cv2.waitKey(1)

58. f.close()

59.  

60. def convertimgset(img_set="train"):

61. imgdir=rootdir+"/WIDER_"+img_set+"/images"

62. gtfilepath=rootdir+"/wider_face_split/wider_face_"+img_set+"_bbx_gt.txt"

63. imagesdir=rootdir+"/images"

64. vocannotationdir=rootdir+"/Annotations"

65. labelsdir=rootdir+"/labels"

66. if not os.path.exists(imagesdir):

67. os.mkdir(imagesdir)

68. if convet2yoloformat:

69. if not os.path.exists(labelsdir):

70. os.mkdir(labelsdir)

71. if convert2vocformat:

72. if not os.path.exists(vocannotationdir):

73. os.mkdir(vocannotationdir)

74. index=0

75. with open(gtfilepath,'r') as gtfile:

76. while(True ):#and len(faces)<10

77. filename=gtfile.readline()[:-1]

78. if(filename==""):

79. break;

80. sys.stdout.write("\r"+str(index)+":"+filename+"\t\t\t")

81. sys.stdout.flush()

82. imgpath=imgdir+"/"+filename

83. img=cv2.imread(imgpath)

84. if not img.data:

85. break;

86. imgheight=img.shape[0]

87. imgwidth=img.shape[1]

88. maxl=max(imgheight,imgwidth)

89. paddingleft=(maxl-imgwidth)>>1

90. paddingright=(maxl-imgwidth)>>1

91. paddingbottom=(maxl-imgheight)>>1

92. paddingtop=(maxl-imgheight)>>1

93. saveimg=cv2.copyMakeBorder(img,paddingtop,paddingbottom,paddingleft,paddingright,cv2.BORDER_CONSTANT,value=0)

94. showimg=saveimg.copy()

95. numbbox=int(gtfile.readline())

96. bboxes=[]

97. for i in range(numbbox):

98. line=gtfile.readline()

99. line=line.split()

100. line=line[0:4]

101. if(int(line[3])<=0 or int(line[2])<=0):

102. continue

103. x=int(line[0])+paddingleft

104. y=int(line[1])+paddingtop

105. width=int(line[2])

106. height=int(line[3])

107. bbox=(x,y,width,height)

108. x2=x+width

109. y2=y+height

110. #face=img[x:x2,y:y2]

111. if width>=minsize2select and height>=minsize2select:

112. bboxes.append(bbox)

113. cv2.rectangle(showimg,(x,y),(x2,y2),(0,255,0))

114. #maxl=max(width,height)

115. #x3=(int)(x+(width-maxl)*0.5)

116. #y3=(int)(y+(height-maxl)*0.5)

117. #x4=(int)(x3+maxl)

118. #y4=(int)(y3+maxl)

119. #cv2.rectangle(img,(x3,y3),(x4,y4),(255,0,0))

120. else:

121. cv2.rectangle(showimg,(x,y),(x2,y2),(0,0,255))

122. filename=filename.replace("/","_")

123. if len(bboxes)==0:

124. print "warrning: no face"

125. continue

126. cv2.imwrite(imagesdir+"/"+filename,saveimg)

127. if convet2yoloformat:

128. height=saveimg.shape[0]

129. width=saveimg.shape[1]

130. txtpath=labelsdir+"/"+filename

131. txtpath=txtpath[:-3]+"txt"

132. ftxt=open(txtpath,'w')

133. for i in range(len(bboxes)):

134. bbox=bboxes[i]

135. xcenter=(bbox[0]+bbox[2]*0.5)/width

136. ycenter=(bbox[1]+bbox[3]*0.5)/height

137. wr=bbox[2]*1.0/width

138. hr=bbox[3]*1.0/height

139. txtline="0 "+str(xcenter)+" "+str(ycenter)+" "+str(wr)+" "+str(hr)+"\n"

140. ftxt.write(txtline)

141. ftxt.close()

142. if convert2vocformat:

143. xmlpath=vocannotationdir+"/"+filename

144. xmlpath=xmlpath[:-3]+"xml"

145. doc = Document()

146. annotation = doc.createElement('annotation')

147. doc.appendChild(annotation)

148. folder = doc.createElement('folder')

149. folder_name = doc.createTextNode('widerface')

150. folder.appendChild(folder_name)

151. annotation.appendChild(folder)

152. filenamenode = doc.createElement('filename')

153. filename_name = doc.createTextNode(filename)

154. filenamenode.appendChild(filename_name)

155. annotation.appendChild(filenamenode)

156. source = doc.createElement('source')

157. annotation.appendChild(source)

158. database = doc.createElement('database')

159. database.appendChild(doc.createTextNode('wider face Database'))

160. source.appendChild(database)

161. annotation_s = doc.createElement('annotation')

162. annotation_s.appendChild(doc.createTextNode('PASCAL VOC2007'))

163. source.appendChild(annotation_s)

164. image = doc.createElement('image')

165. image.appendChild(doc.createTextNode('flickr'))

166. source.appendChild(image)

167. flickrid = doc.createElement('flickrid')

168. flickrid.appendChild(doc.createTextNode('-1'))

169. source.appendChild(flickrid)

170. owner = doc.createElement('owner')

171. annotation.appendChild(owner)

172. flickrid_o = doc.createElement('flickrid')

173. flickrid_o.appendChild(doc.createTextNode('yanyu'))

174. owner.appendChild(flickrid_o)

175. name_o = doc.createElement('name')

176. name_o.appendChild(doc.createTextNode('yanyu'))

177. owner.appendChild(name_o)

178. size = doc.createElement('size')

179. annotation.appendChild(size)

180. width = doc.createElement('width')

181. width.appendChild(doc.createTextNode(str(saveimg.shape[1])))

182. height = doc.createElement('height')

183. height.appendChild(doc.createTextNode(str(saveimg.shape[0])))

184. depth = doc.createElement('depth')

185. depth.appendChild(doc.createTextNode(str(saveimg.shape[2])))

186. size.appendChild(width)

187. size.appendChild(height)

188. size.appendChild(depth)

189. segmented = doc.createElement('segmented')

190. segmented.appendChild(doc.createTextNode('0'))

191. annotation.appendChild(segmented)

192. for i in range(len(bboxes)):

193. bbox=bboxes[i]

194. objects = doc.createElement('object')

195. annotation.appendChild(objects)

196. object_name = doc.createElement('name')

197. object_name.appendChild(doc.createTextNode('face'))

198. objects.appendChild(object_name)

199. pose = doc.createElement('pose')

200. pose.appendChild(doc.createTextNode('Unspecified'))

201. objects.appendChild(pose)

202. truncated = doc.createElement('truncated')

203. truncated.appendChild(doc.createTextNode('1'))

204. objects.appendChild(truncated)

205. difficult = doc.createElement('difficult')

206. difficult.appendChild(doc.createTextNode('0'))

207. objects.appendChild(difficult)

208. bndbox = doc.createElement('bndbox')

209. objects.appendChild(bndbox)

210. xmin = doc.createElement('xmin')

211. xmin.appendChild(doc.createTextNode(str(bbox[0])))

212. bndbox.appendChild(xmin)

213. ymin = doc.createElement('ymin')

214. ymin.appendChild(doc.createTextNode(str(bbox[1])))

215. bndbox.appendChild(ymin)

216. xmax = doc.createElement('xmax')

217. xmax.appendChild(doc.createTextNode(str(bbox[0]+bbox[2])))

218. bndbox.appendChild(xmax)

219. ymax = doc.createElement('ymax')

220. ymax.appendChild(doc.createTextNode(str(bbox[1]+bbox[3])))

221. bndbox.appendChild(ymax)

222. f=open(xmlpath,"w")

223. f.write(doc.toprettyxml(indent = ''))

224. f.close()

225. #cv2.imshow("img",showimg)

226. #cv2.waitKey()

227. index=index+1

228.  

229. def generatetxt(img_set="train"):

230. gtfilepath=rootdir+"/wider_face_split/wider_face_"+img_set+"_bbx_gt.txt"

231. f=open(rootdir+"/"+img_set+".txt","w")

232. with open(gtfilepath,'r') as gtfile:

233. while(True ):#and len(faces)<10

234. filename=gtfile.readline()[:-1]

235. if(filename==""):

236. break;

237. filename=filename.replace("/","_")

238. imgfilepath=datasetprefix+"/images/"+filename

239. f.write(imgfilepath+'\n')

240. numbbox=int(gtfile.readline())

241. for i in range(numbbox):

242. line=gtfile.readline()

243. f.close()

244.  

245. def generatevocsets(img_set="train"):

246. if not os.path.exists(rootdir+"/ImageSets"):

247. os.mkdir(rootdir+"/ImageSets")

248. if not os.path.exists(rootdir+"/ImageSets/Main"):

249. os.mkdir(rootdir+"/ImageSets/Main")

250. gtfilepath=rootdir+"/wider_face_split/wider_face_"+img_set+"_bbx_gt.txt"

251. f=open(rootdir+"/ImageSets/Main/"+img_set+".txt",'w')

252. with open(gtfilepath,'r') as gtfile:

253. while(True ):#and len(faces)<10

254. filename=gtfile.readline()[:-1]

255. if(filename==""):

256. break;

257. filename=filename.replace("/","_")

258. imgfilepath=filename[:-4]

259. f.write(imgfilepath+'\n')

260. numbbox=int(gtfile.readline())

261. for i in range(numbbox):

262. line=gtfile.readline()

263. f.close()

264.  

265. def convertdataset():

266. img_sets=["train","val"]

267. for img_set in img_sets:

268. convertimgset(img_set)

269. generatetxt(img_set)

270. generatevocsets(img_set)

271.  

272. if __name__=="__main__":

273. convertdataset()

274. shutil.move(rootdir+"/"+"train.txt",rootdir+"/"+"trainval.txt")

275. shutil.move(rootdir+"/"+"val.txt",rootdir+"/"+"test.txt")

276. shutil.move(rootdir+"/ImageSets/Main/"+"train.txt",rootdir+"/ImageSets/Main/"+"trainval.txt")

277. shutil.move(rootdir+"/ImageSets/Main/"+"val.txt",rootdir+"/ImageSets/Main/"+"test.txt")

如果没有时间自己转换，也可以下载已经转换好的文件，百度网盘，密码：xsdt