def compareXml(testPath, dstPath, savePath, num_classes=16, conf_thresh=0.3, IOU_thresh=0.3):
"""
计算混淆矩阵
Args:
testPath:预测结果文件
dstPath:真实标注文件
savePath:混淆矩阵保存路径
num_classes:细胞种类
conf_thresh:混淆矩阵置信度阈值
IOU_thresh:混淆矩阵IOU阈值
Returns:
"""
table = np.zeros((num_classes+1, num_classes+1))
for subdir in os.listdir(testPath):
xmllist_test = os.listdir(testPath + subdir)
for xmlfile in xmllist_test:
testfile = testPath + subdir + "/" + xmlfile
dstfile = dstPath + subdir + "/" + xmlfile
DOMTree_test = xml.dom.minidom.parse(testfile) # 打开xml文档
DOMTree_dst = xml.dom.minidom.parse(dstfile)
collection_test = DOMTree_test.documentElement # 得到文档元素对象
collection_dst = DOMTree_dst.documentElement
objectlist_test = collection_test.getElementsByTagName("object") # 得到标签名为object的信息
objectlist_dst = collection_dst.getElementsByTagName("object")
conf_mat = ConfusionMatrix(num_classes=num_classes, CONF_THRESHOLD=conf_thresh, IOU_THRESHOLD=IOU_thresh)
my_result = []
gt_bbox_label = []
for objects_test in objectlist_test:
namelist_test = objects_test.getElementsByTagName('name')
name_test = namelist_test[0].childNodes[0].data
bndbox_test = objects_test.getElementsByTagName('bndbox')
p1 = []
for box in bndbox_test:
p_list = box.getElementsByTagName('xmin')
p1.append(int(p_list[0].childNodes[0].data))
p_list = box.getElementsByTagName('ymin')
p1.append(int(p_list[0].childNodes[0].data))
p_list = box.getElementsByTagName('xmax')
p1.append(int(p_list[0].childNodes[0].data))
p_list = box.getElementsByTagName('ymax')
p1.append(int(p_list[0].childNodes[0].data))
p1.append(1)
p1.append(findIndex(name_test))
my_result.append(p1)
for objects_dst in objectlist_dst:
namelist_dst = objects_dst.getElementsByTagName('name')
name_dst = namelist_dst[0].childNodes[0].data
bndbox_dst = objects_dst.getElementsByTagName('bndbox')
p2 = []
p2.append(findIndex(name_dst))
for box in bndbox_dst:
p_list = box.getElementsByTagName('xmin')
p2.append(int(p_list[0].childNodes[0].data))
p_list = box.getElementsByTagName('ymin')
p2.append(int(p_list[0].childNodes[0].data))
p_list = box.getElementsByTagName('xmax')
p2.append(int(p_list[0].childNodes[0].data))
p_list = box.getElementsByTagName('ymax')
p2.append(int(p_list[0].childNodes[0].data))
gt_bbox_label.append(p2)
my_result = np.array(my_result)
gt_bbox_label = np.array(gt_bbox_label)
if gt_bbox_label.shape == (0,):
continue
# print("my_result = ", my_result)
# print("gt = ", gt_bbox_label)
conf_mat.process_batch(my_result, gt_bbox_label)
tmp = conf_mat.return_matrix()
table = table + tmp
np.savetxt(savePath + "table.txt", table, fmt="%d", delimiter="\t")
def compareXml(testPath, dstPath):
'''比较测试集预测xml和标注的xml，求出混淆矩阵'''
table = np.zeros((17,17))
nums = np.zeros((16,), dtype=np.int16)
for subdir in os.listdir(testPath):
xmllist_test = os.listdir(testPath+subdir)
for xmlfile in xmllist_test:
testfile = testPath + subdir + "/" + xmlfile
dstfile = dstPath + subdir + "/" + xmlfile
DOMTree_test = xml.dom.minidom.parse(testfile) # 打开xml文档
DOMTree_dst = xml.dom.minidom.parse(dstfile)
collection_test = DOMTree_test.documentElement # 得到文档元素对象
collection_dst = DOMTree_dst.documentElement
objectlist_test = collection_test.getElementsByTagName("object") # 得到标签名为object的信息
objectlist_dst = collection_dst.getElementsByTagName("object")
conf_mat = ConfusionMatrix(num_classes=16, CONF_THRESHOLD=0.3, IOU_THRESHOLD=0.3)
my_result = []
gt_bbox_label = []
for objects_test in objectlist_test:
namelist_test = objects_test.getElementsByTagName('name')
name_test = namelist_test[0].childNodes[0].data
bndbox_test = objects_test.getElementsByTagName('bndbox')
p1 = []
nums[findIndex(name_test)] += 1
for box in bndbox_test:
p_list = box.getElementsByTagName('xmin')
p1.append(int(p_list[0].childNodes[0].data))
p_list = box.getElementsByTagName('ymin')
p1.append(int(p_list[0].childNodes[0].data))
p_list = box.getElementsByTagName('xmax')
p1.append(int(p_list[0].childNodes[0].data))
p_list = box.getElementsByTagName('ymax')
p1.append(int(p_list[0].childNodes[0].data))
p1.append(1)
p1.append(findIndex(name_test))
my_result.append(p1)
for objects_dst in objectlist_dst:
namelist_dst = objects_dst.getElementsByTagName('name')
name_dst = namelist_dst[0].childNodes[0].data
if findIndex(name_dst)==15:
continue
bndbox_dst = objects_dst.getElementsByTagName('bndbox')
p2 = []
p2.append(findIndex(name_dst))
for box in bndbox_dst:
p_list = box.getElementsByTagName('xmin')
p2.append(int(p_list[0].childNodes[0].data))
p_list = box.getElementsByTagName('ymin')
p2.append(int(p_list[0].childNodes[0].data))
p_list = box.getElementsByTagName('xmax')
p2.append(int(p_list[0].childNodes[0].data))
p_list = box.getElementsByTagName('ymax')
p2.append(int(p_list[0].childNodes[0].data))
gt_bbox_label.append(p2)
my_result = np.array(my_result)
gt_bbox_label = np.array(gt_bbox_label)
# print("my_result.shape =", my_result.shape)
conf_mat.process_batch(my_result, gt_bbox_label)
tmp = conf_mat.return_matrix()
# input()
table = table + tmp
np.savetxt(savePath+"table.txt", table, fmt="%d", delimiter="\t") # 混淆矩阵
np.savetxt(savePath+"nums.txt", table, fmt="%d", delimiter="\t") # 预测各类细胞个数
return 0