def Test_Dataset(TaskID, datasetBase, datasetConfig, modelConfig, ModelFileName, testData, modelInLogDir = False, showImg = True):
#ROOT_DIR = os.getcwd()
ROOT_DIR = datasetBase + "/MaskRCNN_LOGS"
LOG_DIR = ROOT_DIR + "/logs"
if(modelInLogDir):
MODEL_DIR = LOG_DIR
else:
MODEL_DIR = ROOT_DIR + "/Models"
LOG_DIR_FOLDER = LOG_DIR + "/Test_" + TaskID
Mask_DIR = LOG_DIR_FOLDER + "/Mask"
OVERLAY_DIR = LOG_DIR_FOLDER + "/OverLay"
if(not os.path.isdir(LOG_DIR_FOLDER)):
os.mkdir(LOG_DIR_FOLDER)
if(not os.path.isdir(Mask_DIR)):
os.mkdir(Mask_DIR)
if(not os.path.isdir(OVERLAY_DIR)):
os.mkdir(OVERLAY_DIR)
print("start prepare")
testData.prepare()
testImages, testFileNames, testMasks = testData.getDataSet()
print("load Data Done")
model = modellib.MaskRCNN(mode="inference", model_dir=MODEL_DIR + "/" + ModelFileName, config=modelConfig, log_dir = LOG_DIR)
model.load_weights(MODEL_DIR + "/" + ModelFileName, by_name=True)
print("load weight done")
Classes = datasetConfig.CLASSES
##############################################################
# Run detection
results = model.detect_All(testImages, verbose=1)
results = PostProcessing.GetMostProperMask(results, Classes)
print("Result ! ")
#############################################################
# Estimation
if(testMasks is not None):
dices = []
jaccards = []
ious = []
distances = []
try :
logfile = LOG_DIR_FOLDER + "/Test_" + TaskID + "_Result.csv"
f = open(logfile, 'a')
f_writer = csv.writer(f)
strline = []
strline = ["filename", "Dice", "jaccard", "iou_mean", "dis_min", "dis_max", "dis_mean"]
f_writer.writerow(strline)
for i, result in enumerate(results):
bboxes = result['rois']
if(len(bboxes) == 0):
dices.append(0)
jaccards.append(0)
ious.append(0)
distances.append(0)
strline = []
strline = [testFileNames[i], 0, 0, 0, 0, 0, 0]
f_writer.writerow(strline)
continue
bbox = bboxes[0]
mask = result['masks'][:,:,0]
class_ids = result['class_ids']
truemask = np.asarray(testMasks[i], dtype = "uint8")
ret, truemask = cv2.threshold(truemask, 127, 255, cv2.THRESH_BINARY)
_, contours, _ = cv2.findContours(truemask, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)
true_bbox = cv2.boundingRect(contours[0])
truemask = truemask // 255
mask = np.asarray(mask)
#print(mask.shape, truemask.shape)
dice = Dice(truemask, mask)
jaccard = Jaccard_coef(truemask, mask)
iou = estimate_IoU(true_bbox, (bbox[1], bbox[0], bbox[3] - bbox[1], bbox[2] - bbox[0]))
dis = ave_DisofSkeleton(truemask, mask)
strline = []
strline = [testFileNames[i], dice, jaccard, iou, dis[0], dis[1], dis[2]]
f_writer.writerow(strline)
dices.append(dice)
jaccards.append(jaccard)
ious.append(iou)
distances.append(dis[2])
f.close()
dice = np.asarray(dices).mean()
jaccard = np.asarray(jaccards).mean()
iou = np.asarray(ious).mean()
dis = np.asarray(distances).mean()
except Exception as ex:
print(ex)
if (f):
f.close()
os.rmdir(LOG_DIR_FOLDER)
save_TestHistoryLog(TaskID, ROOT_DIR, ModelFileName, dice, jaccard, iou, dis, datasetConfig)
############################################################
# Display & Save
for i, result in enumerate(results) :
visualize.display_instances(testImages[i], result, Classes, testFileNames[i], auto_show = showImg, truemask =None, logDir = LOG_DIR_FOLDER )
del model
gc.collect()
K.clear_session()
print("Test Done ! ")
