def getImage(num):
# ラズパイで画像が保存されている場所
url = "http://49.135.3.100/~pi/python/images/annotation/"
# ジングルベルはクラス数9個のため、9種類の連番ディレクトリを作成する
for i in range(10):
# 画像の保存フォルダ名
images_dir = "Images/" + "{:03}".format(i + 1)
os.makedirs(images_dir, exist_ok=True)
# 取得画像のurl
imageurl = url + num + ".jpg"
# 元画像を保存するパス
fileOrigin = "Images/original/" + num + ".jpg"
# 変換後の画像を保存する相対パス
filename = make_filename(images_dir, num, imageurl)
# 画像をダウンロード
image = download_image(imageurl)
# 初回だけ元画像を保存
if (i == 0):
os.makedirs("Images/original", exist_ok=True)
save_image(fileOrigin, image)
print("saved original " + num + ".jpg in Images/original/")
#グレイスケールで読み込み
img = cv2.imread(fileOrigin, 0)
# 歪み補正と直線補正
img_tr = image_transform.image_transform(img)
# 適応的2値化
img_two = cv2.adaptiveThreshold(img_tr, 255,
cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
cv2.THRESH_BINARY, 11, 2)
img_two = cv2.cvtColor(img_two, cv2.COLOR_GRAY2RGB)
#切り出し
img_two = img_two[300:600, 200:600]
#２倍に拡大
img_two = cv2.resize(img_two, None, fx=2, fy=2)
cv2.imwrite(filename, img_two)
print("saved two value " + num + ".jpg in " + images_dir)
# -*- coding: utf-8 -*-
import image_transform
import image
import os
import sys
import signal
import cv2
# グレイスケールで読み込む
raw = cv2.imread("slide_image/3.jpg", 0)
cv2.imwrite("slide_image/raw.jpg", raw)
# 歪み補正と直線補正
warp = image_transform.image_transform(raw)
cv2.imwrite("slide_image/warp.jpg", warp)
# 切り出し
cut = warp[300:600, 200:600]
cv2.imwrite("slide_image/cut.jpg", cut)
# ２倍に拡大
enlarge = cv2.resize(cut, None, fx=2, fy=2)
cv2.imwrite("slide_image/enlarge.jpg", enlarge)
# 適応的2値化
binary = cv2.adaptiveThreshold(enlarge, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
cv2.THRESH_BINARY, 11, 2)
binary = cv2.cvtColor(binary, cv2.COLOR_GRAY2RGB)
cv2.imwrite("slide_image/binary.jpg", binary)
if (i == 0):
#動く
move.move()
if (i == N-1):
#止まる
move.stop()
"""
'''
# グレイスケールで読み込む
#img = cv2.imread("images/0.jpg",0)
img = cv2.imread(
"/home/sfc_kamata/work/BBox-Label-Tool/Images/original/test_" +
str(i + 1) + ".jpg", 0)
print('image read %f' % (time.time() - start_time))
# 歪み補正と直線補正
img_tr = image_transform.image_transform(img)
print('transform %f' % (time.time() - start_time))
# 適応的2値化
img_two = cv2.adaptiveThreshold(img_tr, 255,
cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
cv2.THRESH_BINARY, 11, 2)
img_two = cv2.cvtColor(img_two, cv2.COLOR_GRAY2RGB)
print('cvt color %f' % (time.time() - start_time))
# 切り出し
img_two = img_two[300:600, 200:600]
# ２倍に拡大
img_two = cv2.resize(img_two, None, fx=2, fy=2)
# 認識
detectionResults = detectMusic.yoloDetect_net(img_two, net)
print('detection %f' % (time.time() - start_time))
#detectionResults = detectMusic.yoloDetect(img_two)
## saved in GPU -> load in CPU
# load_weights = torch.load(load_path, map_location={"cuda:0": "cpu"})
net.load_state_dict(load_weights)
## input
img_rootpath = "./dataset_lion_tiger"
file_type = "jpg"
val_list = make_datapath_list.make_datapath_list(img_rootpath,
file_type,
phase="val")
size = 224 #VGG16
mean, std = compute_images_mean_std.compute_images_mean_std(
"./dataset_lion_tiger/train", "jpg", resize=size)
transform = image_transform.image_transform(size, mean, std)
## predict(0: lion, 1:tiger)
plt.figure()
i = 0
h = 4
w = 5
for img_path in val_list:
img = Image.open(img_path)
img_trasformed = transform(img)
inputs = img_trasformed.unsqueeze_(0)
outputs = net(inputs)
predict_id = np.argmax(outputs.detach().numpy())
predict_name = ""
print(outputs)
filenames = os.listdir(source_path)
for filename in filenames:
print(filename)
if filename.endswith('0.bmp'):
source_image_path = os.path.join(source_path, filename)
img = cv2.imread(source_image_path) #move below source_image_path to enable file location elsewhere than main
rect = crop.cropping_points2(img)
print(rect)
if not np.any(rect):
print('no save')
os.remove(source_image_path)
continue
# print(rect)
small_image = transform.image_transform(img, dest_path, rect)
# print(small_image_path)
# cv2.imshow('small' , small_image)
answer = simpledialog.askstring("Input", "1: OK\n 2:Seaweed \n 3: Other issue\n 4: Faulty cropping-dont save\n 5:Quit", parent=application_window)
if answer == '5':
exit()
elif answer == '4':
continue
# cv2.waitKey(0)
small_image_name = filename[0:14]+'.'+answer+'.jpg'
small_image_path = os.path.join(dest_path, small_image_name)
os.remove(source_image_path)
print(small_image_path)
print('source: ', source_image_path)
cv2.imshow('small' , small_image)