def get(self, *args, **kwargs):
image = self.get_argument('image', '')
_type = self.get_argument('type', '')
pos = self.get_argument('pos', '')
if not image or not _type or not pos or _type not in ('txt', 'img'):
self.render_obj(dict(code=400, msg='arguments error'))
return
img_image, fp_image, fn_image = utils.get_upload_image_file(
self.upload_path, 'image', image)
if not img_image:
self.render_obj(dict(code=404, msg='image file not found'))
return
wm_path = os.path.join(self.media_path, 'watermark', _type)
if _type == 'img':
mark = self.get_argument('mark', '')
if not mark:
self.render_obj(dict(code=400, msg='args error: mark'))
return
img_mark, fp_mark, _ = utils.get_upload_image_file(
self.upload_path, 'mark', mark)
if not img_mark:
self.render_obj(dict(code=404, msg='mark file not found'))
return
position, img_re_mark = utils.get_mark_position(
img_image, img_mark, pos)
if not position:
self.render_obj(dict(code=404, msg='args error: pos'))
return
fp = '%s-%s' % (fp_image, fp_mark)
elif _type == 'txt':
txt = self.get_argument('txt', '')
font = self.get_argument('font', '')
font_size = int(self.get_argument('fontSize', '24'))
font_color = self.get_argument('fontColor', '')
if not txt or not font or not font_size or not font_color:
self.render_obj(
dict(code=400, msg='args error: txt or font setting'))
return
img_mark = utils.text2img(txt,
os.path.join(self.static_path, 'fonts'),
font,
font_color=font_color,
font_size=font_size)
fp = fp_image
else:
self.render_obj(dict(code=404, msg='args error: type'))
return
position, img_re_mark = utils.get_mark_position(
img_image, img_mark, pos)
if not position:
self.render_obj(dict(code=404, msg='args error: pos'))
return
fn = '%s-%s' % (pos, fn_image)
new_img = utils.img_composite(img_image, img_re_mark, position)
new_img_path = os.path.join(wm_path, fp)
utils.img_save(new_img, new_img_path, fn)
self.render_obj(dict(code=200, msg='success', fp=fp, fn=fn))
def generate_samples(self, c1, c2, args):
c1 = torch.LongTensor([c1]).to(self.device)
c2 = torch.LongTensor([c2]).to(self.device)
c1 = to_onehot(c1, self.device)
c2 = to_onehot(c2, self.device)
#对当前类别随机采样
random_samples = self.generate_class_samples(c1, args.sample_num)
img_save(random_samples, args.random_samples_save_path)
# 生成插值样本
interpolation_samples = self.generate_interpolation_samples(c1, c2, args.sample_num)
img_save(interpolation_samples, args.interpolation_samples_save_path)
def model_test(model, test_iter):
model.eval()
for batch_idx, batch_data in enumerate(test_iter):
batch_size = batch_data[0].shape[0]
batch_data = batch_data[0].to(model.device)
batch_data = batch_data.view(batch_size, -1)
target = batch_data.clone()
_, _, recon_result = model(batch_data)
img_name = 'target' + str(batch_idx)
img_save(target, img_name)
img_name = 'recon_result' + str(batch_idx)
img_save(recon_result, img_name)
def get(self, *args, **kwargs):
try:
app_os = self.get_argument('os', '')
icon = self.get_argument('icon', '')
subscript = self.get_argument('subscript', '')
icon_set = self.app_icon_set.get(app_os)
if not icon_set or not icon:
self.render_obj(dict(code=400, msg='arguments error'))
return
img_icon, fp_icon, _ = utils.get_upload_image_file(
self.upload_path, 'icon', icon)
img_sub, fp_sub = None, ''
if not img_icon:
self.render_obj(dict(code=404, msg='image file not found'))
return
if subscript:
img_sub, fp_sub, _ = utils.get_upload_image_file(
self.upload_path, 'subscript', subscript)
if not img_sub:
self.render_obj(dict(code=404, msg='image file not found'))
return
new_img = utils.img_composite(img_icon,
img_sub) if img_sub else img_icon
fp = '%s-%s' % (fp_icon, fp_sub) if fp_sub else fp_icon
app_os_path = os.path.join(self.media_path, 'icons', app_os)
new_img_path = os.path.join(app_os_path, fp)
for d in icon_set:
size = d.get('size', '')
filename = d.get('filename', '')
if not filename or not size:
continue
d_img = utils.img_resize(new_img, size)
if not d_img:
continue
utils.img_save(d_img, new_img_path, filename)
# 生成所有icon的zip文件
zf_name = os.path.join(app_os_path, '%s.zip' % fp)
zf = zipfile.ZipFile(zf_name, 'w', zipfile.ZIP_DEFLATED)
for fn in os.listdir(new_img_path):
zf.write(os.path.join(new_img_path, fn), os.path.join(fp, fn))
zf.close()
self.render_obj(dict(code=200, msg='success', fp=fp))
except:
traceback.print_exc()
self.render_obj(dict(code=500, msg='system error'))
def generate_samples(self, img1, img2, args):
#随机采样
random_samples = self.generate_random_samples(args.sample_num)
img_save(random_samples, args.random_samples_save_path)
# 生成相似样本
similar_samples = self.generate_similar_samples(
img1.unsqueeze(0), args.sample_num)
img_save(similar_samples, args.similar_samples_save_path)
# 生成插值样本
interpolation_samples = self.generate_interpolation_samples(
img1.unsqueeze(0), img2.unsqueeze(0), args.sample_num)
img_save(interpolation_samples, args.interpolation_samples_save_path)
'''------------------------Data Load-------------------------'''
ref, hr = dataload.dataLoader("./SRNTT1000.h5")
M_t = np.load("autumn1000_M_t.npy")
M_s = np.load("autumn1000_M_s.npy")
train_hr, test_hr, train_ref, test_ref, train_Mt, test_Mt, train_Ms, test_Ms \
= train_test_split(hr, ref, M_t, M_s, test_size=0.2)
train_lr = utils.img_resize(train_hr, 25)
train_lref = utils.img_resize(train_ref, 25)
test_lr = utils.img_resize(test_hr, 25)
test_lred = utils.img_resize(test_ref, 25)
for i in range(test_ref.shape[0]):
path = './result/ref/'+ str(i+1) + '.bmp'
utils.img_save(test_ref[i,:,:,:], path)
path = './result/lr/' + str(i+1) + '.bmp'
utils.img_save(test_lr[i,:,:,:], path)
path = './result/hr/' + str(i+1) + '.bmp'
utils.img_save(test_hr[i,:,:,:], path)
'''----------------------Net Construct-------------------------'''
x = tf.placeholder(tf.float32, [None, 40, 40, 3])
y = tf.placeholder(tf.float32, [None, 160, 160, 3])
Mt_ph= tf.placeholder(tf.float32, [None, 40, 40, 256])
Ms_ph= tf.placeholder(tf.float32, [None, 40, 40, 256])
train_mode = tf.placeholder(tf.bool)
Learning_rate = tf.placeholder(tf.float32, shape=[])
ce_net = ce_model.CE(x, train_mode)
k = 0
for i in range(65):
path = '../dataset/91-image/t' + str(i + 1) + '.bmp'
img = utils.img_read(path)
img = utils.img_crop(img, 160, 160)
if img.shape != (160, 160, 3):
continue
train_hr[k, :, :, :] = img
img = utils.img_downsize(img, 25)
train_lr[k, :, :, :] = img
k += 1
k = 0
for i in range(26):
path = '../dataset/91-image/tt' + str(i + 1) + '.bmp'
img = utils.img_read(path)
img = utils.img_crop(img, 160, 160)
if img.shape != (160, 160, 3):
continue
test_hr[k, :, :, :] = img
img = utils.img_downsize(img, 25)
test_lr[k, :, :, :] = img
k += 1
prob = test_ce(train_lr, train_hr, test_lr, test_hr)
for i in range(10):
path = '../dataset/91-image/t' + str(i + 1) + '_.bmp'
utils.img_save(prob[i, :, :, :], path)