"""
This is the code for analyzing detection results, calculating patient score. Generate .txt for all the bbox labels respectively and name them by patientID_frameId.txt
Format: Ymin Xmin Ymax Xmax
eg., 361.0 293.0 397.0 325.0
file name example: 48_12.txt
@detect_res_file: the txt file contains the detection results. Format: jpg name,Ymin,Xmin,Ymax,Xmax,score
jpg name format:patientId_frameId.jpg
eg: 18_10.jpg,491.96744,113.14628,533.2568,153.23444,0.9348341,0.7855689961758875
@test_id: the tested patient or you wanna test all patients, input[]
eg:[18,48,60]
The output is written into eva_result.txt: format: patient id(missing if all patients are tested), patient score
eg.18,1.0
"""
import numpy as np
detect_res_file = "crowd_vote_result.txt"
patient_ids=[18]
def main(detect_res_file, patient_ids):
if len(patient_ids)==0:
return(str(patient_score_eva(detect_res_file,-1)))
else:
score=list()
for patient_id in patient_ids:
cur_score = str(patient_score_eva(detect_res_file,patient_id))
score.append(str(patient_id)+","+cur_score)
return score
#this is the function for calculating iou
def bbox_iou(a, b):
epsilon = 1e-5
# COORDINATES OF THE INTERSECTION BOX
y1 = max(a[0], b[0])
x1 = max(a[1], b[1])
y2 = min(a[2], b[2])
x2 = min(a[3], b[3])
# AREA OF OVERLAP - Area where the boxes intersect
width = (x2 - x1)
height = (y2 - y1)
# handle case where there is NO overlap
if (width < 0) or (height < 0):
return 0.0
area_overlap = width * height
# COMBINED AREA
area_a = (a[2] - a[0]) * (a[3] - a[1])
area_b = (b[2] - b[0]) * (b[3] - b[1])
area_combined = area_a + area_b - area_overlap
# RATIO OF AREA OF OVERLAP OVER COMBINED AREA
iou = area_overlap / (area_combined+epsilon)
return iou
#this is the function for calculating patient score
def patient_score_eva(file_name,patient_id):
with open(file_name,'r') as f:
lines = f.readlines()
info=list()
img_name=list()
label_boxes = dict()
for line in lines:
cur_patient = int(line.split("_")[0])
if cur_patient==patient_id or patient_id==-1:
sp=line.split("jpg")
cur_img_name = sp[0]
cur_frame = int(cur_img_name.split("_")[0].replace(".",""))
cur_box_info = sp[1].split(",")
#find the label for cur img
if cur_img_name in label_boxes:
label_box=label_boxes[cur_img_name]
else:
cur_img_label = cur_img_name.replace(".",".txt")
label_box=[-1,-1,-1,-1]
with open(cur_img_label) as f:
cur_content = f.read().strip('\n')
ls_label_box = cur_content.split(" ")
for i in range(4):
label_box[i]=float(ls_label_box[i])
label_boxes[cur_img_name]=label_box
#read loc for cur img and calculate iou
cur_box=[-1,-1,-1,-1]
for i in range(4):
cur_box[i]=float(cur_box_info[i+1])
cur_iou = bbox_iou(cur_box,label_box)
#read score for cur img
cur_score = float(cur_box_info[5].replace("\n",""))
cur_info=[cur_patient,cur_frame,cur_iou,cur_score]
info.append(cur_info)
img_name.append(cur_img_name)
m=np.array(info)
select_res=list()
#if one frame has several bounding boxes, then selecet the one with max score
for cur_img_name in set(img_name):
cur_patient = int(cur_img_name.split("_")[0])
cur_frame = int(cur_img_name.split("_")[0].replace(".",""))
max_score=-1
select_img = -1
for i in range(len(m)):
if m[i,0]==cur_patient and m[i,1]==cur_frame and m[i,3]>max_score:
max_score=m[i,1]
select_img=i
select_res.append(m[select_img,])
#find the number of res with iou>0
r=np.array(select_res)
count=0
for i in range(len(r)):
if r[i,2]>0:
count+=1
patient_score = count/len(r)
return patient_score
if __name__ == '__main__':
eva_result=main(detect_res_file, patient_ids)
with open('eva_result.txt', 'w') as f:
for item in eva_result:
f.write("%s" % item)
Datasets
Models
