def predict_add(img_path):
height, width, channels = 36, 60, 4
sum = 0
img_processed = preprocess_img(img_path)
if not img_processed:
return None
for i, img_array in enumerate(img_processed):
print(img_array.shape)
img_array = img_array.reshape(1, img_array.shape[0], width,
channels)
value = predict_img(img_array, model=mode[i], mapp=cls_map[i])
sum += value
return sum
def save_img():
file = request.files['image']
file.save(os.path.join(app.config['UPLOAD_FOLDER'], file.filename))
result = predict_img(file.filename)
print(result)
return result
in_file = args.test + file_name + "/images/" + file_name + ".png"
out_file = args.save + file_name + ".png"
print("\nPredicting image {} ...".format(in_file))
original_img = Image.open(in_file)
# 元画像の大きさを保存しておく
original_width = original_img.size[0]
original_height = original_img.size[1]
original_img = np.asarray(original_img)
# 所定の大きさにresize
resized_img = to_square(original_img, args.size)
# U-Netを用いてsegentation（確率値を出力）
probs = predict_img(net, resized_img, args.gpu)
# もとの大きさに戻す
probs_resized = from_square(probs, (original_height, original_width))
# ノイズ除去 & 二値化
dst_img = remove_noise(probs_resized,
(original_height, original_width), original_img)
dst_img = (dst_img * 255).astype(np.uint8)
# 真ん中を穴埋め
if "b83d1d77935b6cfd44105b54600ffc4b6bd82de57dec65571bcb117fa8398ba3" in in_file:
print("zzzzzzzzzzzzzzzzzz")
else:
dst_img = fill(dst_img)
out_file = args.save + file_name + ".png"
print("\nPredicting image {} ...".format(in_file))
original_img = Image.open(in_file)
original_width = original_img.size[0]
original_height = original_img.size[1]
original_img_array = np.asarray(original_img)
# 染色方法を識別
image_type = imagetype_classification(in_file)
resized_img = original_img.resize(SIZE)
#resized_img = original_img
resized_img_array = np.array(resized_img)
print(image_type, "\n")
dst_img_array = predict_img(net, resized_img_array, args.gpu)
dst_img_array = morphology(dst_img_array, iterations=1)
dst_img = Image.fromarray(dst_img_array * 255)
dst_img_resized = dst_img.resize((original_width, original_height))
print(original_width, original_height)
#dst_img_resized_array = np.asarray(dst_img_resized)
#devide = Devide(original_img_array, dst_img_resized_array)
#out = devide.make_mask()
#result = Image.fromarray((out).astype(numpy.uint8))
result = dst_img_resized
result.save(out_file)
for file_name in os.listdir(args.test):
in_file = args.test + file_name + "/images/" + file_name + ".png"
out_file = args.save + file_name + ".png"
print("\nPredicting image {} ...".format(in_file))
img = Image.open(in_file)
original_width = img.size[0]
original_height = img.size[1]
img_array = np.asarray(img)
# 染色方法ごとに処理を分ける
image_type = imagetype_classification(in_file)
print(image_type, "\n")
if image_type == 1:
out = predict_img(net, img, in_file, args.gpu, SIZE)
out = morphology(out, iterations=0)
result = Image.fromarray((out * 255).astype(numpy.uint8))
result = result.resize((original_width, original_height))
elif image_type == 2:
out = predict_img(net, img, in_file, args.gpu, SIZE)
out = morphology(out, iterations=0)
result = Image.fromarray((out * 255).astype(numpy.uint8))
result = result.resize((original_width, original_height))
else:
out = predict_img(net, img, in_file, args.gpu, SIZE)
out = morphology(out, iterations=0)
result = Image.fromarray((out * 255).astype(numpy.uint8))
result = result.resize((original_width, original_height))
result.save(out_file)