def visual_patch(self):
gen = self.train_gen()
(X, Y) = next(gen)
image = []
mask = []
print("[INFO] Visualize Image Sample...")
for index in range(self.config.batch_size):
image.append(X[index])
mask.append(
np.reshape(Y, [
self.config.batch_size, self.config.patch_height,
self.config.patch_width, self.config.seg_num + 1
])[index, :, :, 0])
if self.config.batch_size % 4 == 0:
row = self.config.batch_size / 4
col = 4
else:
if self.config.batch_size % 5 != 0:
row = self.config.batch_size // 5 + 1
else:
row = self.config.batch_size // 5
col = 5
imagePatch = visualize(image, [row, col])
maskPatch = visualize(mask, [row, col])
cv2.imwrite(self.config.checkpoint + "image_patch.jpg", imagePatch)
cv2.imwrite(self.config.checkpoint + "groundtruth_patch.jpg",
maskPatch)
def predict(self):
predList = glob.glob(self.config.test_img_path + "*." +
self.config.test_datatype)
for path in predList:
orgImg = plt.imread(path)
print("[Info] 解析文件名...", self.analyze_name(path))
height, width = orgImg.shape[:2]
patches_pred, new_height, new_width, adjustImg = get_test_patches(
orgImg, self.config)
predictions = self.model.predict(patches_pred,
batch_size=32,
verbose=1)
pred_patches = pred_to_patches(predictions, self.config)
pred_imgs = recompone_overlap(pred_patches, self.config,
new_height, new_width)
pred_imgs = pred_imgs[:, 0:height, 0:width, :]
adjustImg = adjustImg[0, 0:height, 0:width, :]
print(adjustImg.shape)
probResult = pred_imgs[0, :, :, 0]
binaryResult = gray2binary(probResult)
resultMerge = visualize([adjustImg, binaryResult], [1, 2])
resultMerge = cv2.cvtColor(resultMerge, cv2.COLOR_RGB2BGR)
cv2.imwrite(
self.config.test_result_path + self.analyze_name(path) +
"_merge.jpg", resultMerge)
cv2.imwrite(
self.config.test_result_path + self.analyze_name(path) +
"_prob.jpg", (probResult * 255).astype(np.uint8))
def predict(self):
predList = glob.glob(self.config.test_img_path + "*." +
self.config.test_datatype)
print(len(predList))
for path in predList:
orgImg_temp = plt.imread(path)
#orgImg_temp=cv2.resize(orgImg_temp,(1727,1168))
# orgImg=orgImg_temp[:,:,1]
thresh = 0.6
print('G_ratio:{} R_ratio:{}'.format(thresh, (1 - thresh)))
orgImg = orgImg_temp[:, :, 1] * thresh + orgImg_temp[:, :, 0] * (
1 - thresh)
print("[Info] Analyze filename...", self.analyze_name(path))
height, width = orgImg.shape[:2]
orgImg = np.reshape(orgImg, (height, width, 1))
patches_pred, new_height, new_width, adjustImg = get_test_patches(
orgImg, self.config)
predictions = self.model.predict(patches_pred,
batch_size=32,
verbose=1)
# print(predictions.shape)
pred_patches = pred_to_patches(predictions, self.config)
pred_imgs = recompone_overlap(pred_patches, self.config,
new_height, new_width)
# print(pred_imgs.shape)
gc.collect()
pred_imgs = pred_imgs[:, 0:height, 0:width, :]
new_construct = np.zeros(
(pred_imgs.shape[1], pred_imgs.shape[2], 3))
for i in range(pred_imgs.shape[1]):
for j in range(pred_imgs.shape[2]):
# print(pred_imgs[0][i][j])
index = np.argmax(pred_imgs[0][i][j])
# print(index)
new_construct[i][j] = color_list[index]
# print(pred_imgs.shape)
adjustImg = adjustImg[0, 0:height, 0:width, :]
# print(adjustImg.shape)
probResult = pred_imgs[0, :, :,
0] + pred_imgs[0, :, :,
2] + pred_imgs[0, :, :, 3]
binaryResult = gray2binary(probResult, 0.35)
resultMerge = visualize([adjustImg, binaryResult], [1, 2])
resultMerge = cv2.cvtColor(resultMerge, cv2.COLOR_RGB2BGR)
#cv2.imwrite(self.config.test_result_path+self.analyze_name(path)+"_merge.jpg",resultMerge)
cv2.imwrite(
self.config.test_result_path + self.analyze_name(path) +
"_prob.tif", ((1 - probResult) * 255).astype(np.uint8))
cv2.imwrite(
self.config.test_result_path + self.analyze_name(path) +
"_color_mask.tif", (new_construct).astype(np.uint8))
def predict(self):
predList = glob.glob(self.config.test_img_path + "*." +
self.config.test_datatype)
# probList=glob.glob(self.config.test_img_path+"*."+self.config.test_datatype)
# print(len(predList))
for path in predList:
orgImg_temp = plt.imread(path)
prob_path = self.config.test_prob_path + path.split(
'/')[1].replace('test', 'test_prob')
probImg = plt.imread(prob_path, 0)
#orgImg_temp=cv2.resize(orgImg_temp,(1727,1168))
# orgImg=orgImg_temp[:,:,1]
thresh = 1
print('G_ratio:{} R_ratio:{}'.format(thresh, (1 - thresh)))
orgImg = orgImg_temp[:, :, 1] * thresh + orgImg_temp[:, :, 0] * (
1 - thresh)
print("[Info] Analyze filename...", self.analyze_name(path))
height, width = orgImg.shape[:2]
orgImg = np.reshape(orgImg, (height, width, 1))
probImg = np.reshape(probImg, (height, width, 1))
patches_pred, new_height, new_width, adjustImg = get_test_patches1(
orgImg, probImg, self.config)
predictions = self.model.predict(patches_pred,
batch_size=32,
verbose=1)
pred_patches = pred_to_patches(predictions, self.config)
pred_imgs = recompone_overlap(pred_patches, self.config,
new_height, new_width)
gc.collect()
pred_imgs = pred_imgs[:, 0:height, 0:width, :]
adjustImg = adjustImg[0, 0:height, 0:width, :]
print(adjustImg.shape)
probResult = pred_imgs[0, :, :, 0]
binaryResult = gray2binary(probResult, 0.35)
resultMerge = visualize([adjustImg, binaryResult], [1, 2])
resultMerge = cv2.cvtColor(resultMerge, cv2.COLOR_RGB2BGR)
cv2.imwrite(
self.config.test_result_path + self.analyze_name(path) +
"_merge.jpg", resultMerge)
cv2.imwrite(
self.config.test_result_path + self.analyze_name(path) +
"_prob.bmp", (probResult * 255).astype(np.uint8))
cv2.imwrite(
self.config.test_result_path +
self.analyze_name(path).replace('training', 'manual1') +
".tif", (binaryResult * 255).astype(np.uint8))
def predict(self):
predList = glob.glob(self.config.test_img_path + "*." +
self.config.test_datatype)
print(len(predList))
for index, path in enumerate(predList):
orgImg_temp = plt.imread(predList[index])
# orgImg=orgImg_temp[:,:,1]
thresh = 0.8
print('G_ratio:{} R_ratio:{}'.format(thresh, (1 - thresh)))
orgImg = orgImg_temp[:, :, 1] * thresh + orgImg_temp[:, :, 0] * (
1 - thresh)
print("[Info] Analyze filename...", self.analyze_name(path))
height, width = orgImg.shape[:2]
orgImg = np.reshape(orgImg, (height, width, 1))
patches_pred, new_height, new_width, adjustImg = get_test_patches(
orgImg, self.config)
# self.model.load_weights('./metric/STARE/weights/STARE_1_weight/keras_{}.model'.format(index))
self.model.load_weights(
'./metric/STARE/Unetweights/0.9/keras_{}.model'.format(index))
print("[Info] loading weight....{}".format(index))
predictions = self.model.predict(patches_pred,
batch_size=32,
verbose=1)
pred_patches = pred_to_patches(predictions, self.config)
pred_imgs = recompone_overlap(pred_patches, self.config,
new_height, new_width)
gc.collect()
pred_imgs = pred_imgs[:, 0:height, 0:width, :]
adjustImg = adjustImg[0, 0:height, 0:width, :]
print(adjustImg.shape)
probResult = pred_imgs[0, :, :, 0]
binaryResult = gray2binary(probResult)
resultMerge = visualize([adjustImg, binaryResult], [1, 2])
resultMerge = cv2.cvtColor(resultMerge, cv2.COLOR_RGB2BGR)
cv2.imwrite(
self.config.test_result_path + self.analyze_name(path) +
"_merge.jpg", resultMerge)
cv2.imwrite(
self.config.test_result_path + self.analyze_name(path) +
"_prob.bmp", (probResult * 255).astype(np.uint8))
cv2.imwrite(
self.config.test_result_path + self.analyze_name(path) +
".tif", (binaryResult * 255).astype(np.uint8))
np.save(
self.config.test_result_path + self.analyze_name(path) +
".npy", binaryResult)
def predict(self):
predList = glob.glob(self.config.test_img_path + "*." +
self.config.test_datatype)
for path in predList:
print(path)
baseName = os.path.basename(path)
orgImg_temp = cv2.imread(path)[..., 0]
#orgImg=orgImg_temp[:,:,1]*0.75+orgImg_temp[:,:,0]*0.25
orgImg = orgImg_temp
print("[Info] Analyze filename...", self.analyze_name(path))
height, width = orgImg.shape[:2]
orgImg = np.reshape(orgImg, (height, width, 1))
patches_pred, new_height, new_width, adjustImg = get_test_patches(
orgImg, self.config)
predictions = self.model.predict(patches_pred,
batch_size=32,
verbose=1)
pred_patches = pred_to_patches(predictions, self.config)
pred_imgs = recompone_overlap(pred_patches, self.config,
new_height, new_width)
pred_imgs = pred_imgs[:, 0:height, 0:width, :]
adjustImg = adjustImg[0, 0:height, 0:width, :]
print(adjustImg.shape)
probResult = pred_imgs[0, :, :, 0]
binaryResult = gray2binary(probResult)
resultMerge = visualize([adjustImg, binaryResult], [1, 2])
resultMerge = cv2.cvtColor(resultMerge, cv2.COLOR_RGB2BGR)
test_path = os.path.join(self.config.test_result_path,
self.analyze_name(path),
self.config.exp_name,
self.config.preprocess,
self.config.dataset, 'result')
#test_label = os.path.join(self.config.test_result_path, self.analyze_name(path), self.config.exp_name, self.config.preprocess, self.config.dataset, 'labels')
mkdir_if_not_exist(test_path)
#mkdir_if_not_exist(test_label)
#cv2.imwrite(self.config.test_result_path+self.analyze_name(path)+"test.png", orgImg)
#cv2.imwrite(self.config.test_result_path+self.analyze_name(path)+"_merge.png",resultMerge)
cv2.imwrite(os.path.join(test_path, baseName),
(probResult * 255).astype(np.uint8))
def predict(self):
predList = glob.glob(self.config.test_img_path + "*." +
self.config.test_datatype)
for path in predList:
orgImg_temp = plt.imread(path)
if orgImg_temp.shape[0] > 1000 or orgImg_temp.shape[1] > 1000:
orgImg_temp = imgResize(orgImg_temp, 0.5)
orgImg = orgImg_temp[:, :, 1] * 0.75 + orgImg_temp[:, :, 0] * 0.25
print("[Info] Analyze filename...", self.analyze_name(path))
height, width = orgImg.shape[:2]
orgImg = np.reshape(orgImg, (height, width, 1))
patches_pred, new_height, new_width, adjustImg = get_test_patches(
orgImg, self.config)
predictions = self.model.predict(patches_pred,
batch_size=32,
verbose=1)
pred_patches = pred_to_patches(predictions, self.config)
pred_imgs = recompone_overlap(pred_patches, self.config,
new_height, new_width)
pred_imgs = pred_imgs[:, 0:height, 0:width, :]
probResult = pred_imgs[0, :, :, 0]
binaryResult = gray2binary(probResult, threshold=0.5)
binaryResult, probResult = postprocess(
binaryResult, probResult) # post process to get the optic disc
_, contours, _ = cv2.findContours(binaryResult, cv2.RETR_TREE,
cv2.CHAIN_APPROX_SIMPLE)
adjustImg = cv2.drawContours(orgImg_temp, contours, -1,
(0, 255, 0), 2)
resultMerge = visualize([adjustImg / 255., binaryResult], [1, 2])
resultMerge = cv2.cvtColor(resultMerge, cv2.COLOR_RGB2BGR)
cv2.imwrite(
self.config.test_result_path + self.analyze_name(path) +
"_merge.jpg", resultMerge)
cv2.imwrite(
self.config.test_result_path + self.analyze_name(path) +
"_prob.bmp", (probResult * 255).astype(np.uint8))
def predict(self):
predList = glob.glob(self.config.test_img_path + "*." +
self.config.test_datatype)
print(predList)
for path in predList:
orgImg_temp = plt.imread(path)
# img = Image.open(path)
# img = img.resize((370, 250))
# orgImg_temp = np.asarray(img)
orgImg = orgImg_temp[:, :, 1] * 0.75 + orgImg_temp[:, :, 0] * 0.25
print("[Info] Analyze filename...", self.analyze_name(path))
height, width = orgImg.shape[:2]
orgImg = np.reshape(orgImg, (height, width, 1))
patches_pred, new_height, new_width, adjustImg = get_test_patches(
orgImg, self.config)
predictions = self.model.predict(patches_pred,
batch_size=32,
verbose=1)
pred_patches = pred_to_patches(predictions, self.config)
pred_imgs = recompone_overlap(pred_patches, self.config,
new_height, new_width)
pred_imgs = pred_imgs[:, 0:height, 0:width, :]
adjustImg = adjustImg[0, 0:height, 0:width, :]
print(adjustImg.shape)
probResult = pred_imgs[0, :, :, 0]
binaryResult = gray2binary(probResult)
resultMerge = visualize([adjustImg, binaryResult], [1, 2])
resultMerge = cv2.cvtColor(resultMerge, cv2.COLOR_RGB2BGR)
cv2.imwrite(
self.config.test_result_path + self.analyze_name(path) +
"_merge.jpg", resultMerge)
cv2.imwrite(
self.config.test_result_path + self.analyze_name(path) +
"_prob.bmp", (probResult * 255).astype(np.uint8))
