def test(self):
if self.load_model():
print(' [*] Load SUCCESS!')
else:
print(' [!] Load Failed...')
# read test data
test_data_files = utils.all_files_under(self.dataset.night_path)
total_time = 0.
for idx in range(len(test_data_files)):
img = utils.imagefiles2arrs([test_data_files[idx]]) # read img
img = utils.transform(img) # convert [0, 255] to [-1., 1.]
# measure inference time
start_time = time.time()
imgs = self.model.test_step(img, mode='YtoX') # inference
total_time += time.time() - start_time
self.model.plots(imgs, idx, self.dataset.image_size,
self.test_out_dir) # write results
print('Avg PT: {:3f} msec.'.format(total_time / len(test_data_files) *
1000.))
# load the model and weights
with open(f_model.format(dataset), 'r') as f:
model = model_from_json(f.read())
model.load_weights(f_weights.format(dataset))
# iterate all images
img_size = (640, 640) if dataset == "DRIVE" else (720, 720)
ori_shape = (1, 584, 565) if dataset == "DRIVE" else (1, 605,
700) # batchsize=1
fundus_files = utils.all_files_under(fundus_dir.format(dataset))
mask_files = utils.all_files_under(mask_dir.format(dataset))
for index, fundus_file in enumerate(fundus_files):
print("processing {}...".format(fundus_file))
# load imgs
img = utils.imagefiles2arrs([fundus_file])
mask = utils.imagefiles2arrs([mask_files[index]])
# z score with mean, std (batchsize=1)
mean = np.mean(img[0, ...][mask[0, ...] == 255.0], axis=0)
std = np.std(img[0, ...][mask[0, ...] == 255.0], axis=0)
img[0, ...] = (img[0, ...] - mean) / std
# run inference
padded_img = utils.pad_imgs(img, img_size)
#padded_img=utils.rescale_imgs(img, img_size)
vessel_img = model.predict(padded_img, batch_size=1) * 255
# rescaled_vessel = utils.rescale_to_original(vessel_img[...,0], ori_shape)
# final_result = utils.remain_in_mask(rescaled_vessel[0,...], mask[0,...])
cropped_vessel = utils.crop_to_original(vessel_img[..., 0], ori_shape)
final_result = utils.remain_in_mask(cropped_vessel[0, ...], mask[0,
)
FLAGS, _ = parser.parse_known_args()
os.environ['CUDA_VISIBLE_DEVICES'] = FLAGS.gpu_index
# load the models and corresponding weights
with open("../model/vessel/network.json", 'r') as f:
vessel_model = model_from_json(f.read())
vessel_model.load_weights("../model/vessel/network_weight.h5")
# make directories
if not os.path.isdir(FLAGS.out_dir):
os.makedirs(FLAGS.out_dir)
# iterate all images
img_size = (640, 640)
filenames = utils.all_files_under(FLAGS.img_dir)
for filename in filenames:
# load an img (tensor shape of [1,h,w,3])
img = utils.imagefiles2arrs([filename])
_, h, w, _ = img.shape
# z score with mean, std (batchsize=1)
mean = np.mean(img[0, ...][img[0, ..., 0] > 40.0], axis=0)
std = np.std(img[0, ...][img[0, ..., 0] > 40.0], axis=0)
img[0, ...] = (img[0, ...] - mean) / std
assert len(mean) == 3 and len(std) == 3
# run inference & save the result
vessel = vessel_model.predict(img, batch_size=1)
Image.fromarray((vessel[0, ..., 0]*255).astype(np.uint8)).save(os.path.join(FLAGS.out_dir, os.path.basename(filename)))
import utils
import os
import numpy as np
from sklearn.metrics.classification import confusion_matrix
import argparse
from skimage import measure
from scipy.spatial.distance import dice
from keras.models import model_from_json
os.environ['CUDA_VISIBLE_DEVICES'] = "3"
gt_dir = "../data/original/OD_Segmentation_Training_Set/"
gt_filenames = utils.all_files_under(gt_dir)
gt = utils.imagefiles2arrs(gt_filenames).astype(np.uint8)
# load the models and corresponding weights
with open("../model/vessel/network.json", 'r') as f:
vessel_model = model_from_json(f.read())
vessel_model.load_weights("../model/vessel/network_weight.h5")
with open("../model/od_from_fundus_vessel/network.json", 'r') as f:
od_from_fundus_vessel_model = model_from_json(f.read())
od_from_fundus_vessel_model.load_weights(
"../model/od_from_fundus_vessel/network_weight.h5")
# iterate all images
filenames = utils.all_files_under("../data/original/Training_c/")
pred = np.zeros(gt.shape)
for index, filename in enumerate(filenames):
assert "IDRiD_" in os.path.basename(filename)
# load an img (tensor shape of [1,h,w,3])
img = utils.imagefiles2arrs([filename])
for model_type in ["final_model"]:
for task in tasks:
if task == "SE":
val_indices = [49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61]
else:
val_indices = [0, 1, 2, 3, 4, 54, 55, 56, 57, 58, 59, 60, 61]
gt_dir = gt_dir_template.format(task)
pred_dir = pred_dir_template.format(model_type, task)
gt_filenames = utils.all_files_under(gt_dir)
pred_filenames = utils.all_files_under(pred_dir)
training_indices = ~np.in1d(range(len(pred_filenames)), val_indices)
# build gt arrays
index_gt = 0
pred_all = utils.imagefiles2arrs(pred_filenames)
gt_all = np.zeros(pred_all.shape)
for index_pred in range(len(pred_filenames)):
# build array of gt
if index_gt < len(gt_filenames) and os.path.basename(
pred_filenames[index_pred]).replace(
".jpg", "") in os.path.basename(
gt_filenames[index_gt]):
gt = utils.imagefiles2arrs(
gt_filenames[index_gt:index_gt + 1]).astype(np.uint8)[0,
...]
gt_all[index_pred, ...] = gt
index_gt += 1
# compute sensitivity and specificity
aupr_all, best_f1_all, best_f1_thresh_all, sen_all, ppv_all = utils.pr_metric(
import utils
import os
import numpy as np
from sklearn.metrics.classification import confusion_matrix
import argparse
from skimage import measure
from scipy.spatial.distance import dice
gt_dir = "../data/original/OD_Segmentation_Training_Set/"
pred_dir = "../../inference_codes/outputs_sub3/segmentation/"
gt_filenames = utils.all_files_under(gt_dir)
pred_filenames = utils.all_files_under(pred_dir)
list_sen, list_spe, list_jaccard, list_dice, list_dice_ori = [], [], [], [], []
for index in range(len(gt_filenames)):
# get gt and pred segs
gt = utils.imagefiles2arrs(gt_filenames[index:index + 1]).astype(
np.uint8)[0, ...]
pred = utils.imagefiles2arrs(pred_filenames[index:index + 1]).astype(
np.uint8)[0, ...] // 255
assert len(gt.shape) == 2 and len(pred.shape) == 2
# compute sensitivity and specificity
print pred_filenames[index]
cm, spe, sen, dice_val, jaccard_val = utils.seg_metrics(gt, pred)
# print results and store to lists
print "--- {} ---".format(os.path.basename(gt_filenames[index]))
print "specificity: {}".format(spe)
print "sensitivity: {}".format(sen)
print "jaccard: {}".format(jaccard_val)
print "dice: {}".format(dice_val)
list_spe.append(spe)
import utils
import argparse
# arrange arguments
parser = argparse.ArgumentParser()
FLAGS, _ = parser.parse_known_args()
gt_dir = "../data/original/OD_Segmentation_Training_Set/"
pred_dir = "../outputs/final_results_loss_weight_1_0.1_non_thre/od_from_fundus_vessel_visualization"
gt_filenames = utils.all_files_under(gt_dir)
pred_filenames = utils.all_files_under(pred_dir)
# build gt arrays
pred_all = utils.imagefiles2arrs(pred_filenames)
pred_all /= 255.
gt_all = utils.imagefiles2arrs(gt_filenames)
dice, threshold = utils.segmentation_optimal_threshold(gt_all, pred_all)
print "dice: {}".format(dice)
print "best threshold: {}".format(threshold)
localization_model = model_from_json(f.read())
localization_model.load_weights(
"../model/od_fovea_localization/network_weight.h5")
# make directories
img_out_dir = os.path.join(FLAGS.out_dir, "pts_check")
if not os.path.isdir(img_out_dir):
os.makedirs(img_out_dir)
# iterate all images
filepaths = utils.all_files_under(FLAGS.img_dir)
image_no, od, fovea = [], {"x": [], "y": []}, {"x": [], "y": []}
for filepath in filepaths:
assert "IDRiD_" in os.path.basename(filepath)
# load an img (tensor shape of [1,h,w,3])
img = utils.imagefiles2arrs([filepath])
_, h, w, _ = img.shape
assert h == 2848 and w == 4288
resized_img = utils.resize_img(img)
# run inference
vessel = vessel_model.predict(utils.normalize(resized_img,
"vessel_segmentation"),
batch_size=1)
predicted_coords, _ = localization_model.predict([
utils.normalize(resized_img, "od_from_fundus_vessel"),
(vessel * 255).astype(np.uint8) / 255.
],
batch_size=1)
# convert to coordinates in the original image