def test_video_truth(self, path, name='result', reuse=False, part=50):
save_path=join(path,name)
automkdir(save_path)
inp_path=join(path,'truth')
imgs=sorted(glob.glob(join(inp_path,'*.png')))
max_frame=len(imgs)
imgs=np.array([cv2_imread(i) for i in imgs])/255.
if part>max_frame:
part=max_frame
if max_frame%part ==0 :
num_once=max_frame//part
else:
num_once=max_frame//part+1
h,w,c=imgs[0].shape
L_test = tf.placeholder(tf.float32, shape=[num_once, self.num_frames, h//self.scale, w//self.scale, 3], name='L_test')
SR_test=self.forward(L_test)
if not reuse:
self.img_hr=tf.placeholder(tf.float32, shape=[None, None, None, 3], name='H_truth')
self.img_lr=DownSample_4D(self.img_hr, BLUR, scale=self.scale)
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
#sess=tf.Session()
self.sess=sess
sess.run(tf.global_variables_initializer())
self.saver = tf.train.Saver(max_to_keep=100, keep_checkpoint_every_n_hours=1)
self.load(sess, self.save_dir)
lrs=self.sess.run(self.img_lr,feed_dict={self.img_hr:imgs})
lr_list=[]
max_frame=lrs.shape[0]
for i in range(max_frame):
index=np.array([i for i in range(i-self.num_frames//2,i+self.num_frames//2+1)])
index=np.clip(index,0,max_frame-1).tolist()
lr_list.append(np.array([lrs[j] for j in index]))
lr_list=np.array(lr_list)
print('Save at {}'.format(save_path))
print('{} Inputs With Shape {}'.format(lrs.shape[0],lrs.shape[1:]))
h,w,c=lrs.shape[1:]
all_time=[]
for i in trange(part):
st_time=time.time()
sr=self.sess.run(SR_test,feed_dict={L_test : lr_list[i*num_once:(i+1)*num_once]})
all_time.append(time.time()-st_time)
for j in range(sr.shape[0]):
img=sr[j][0]*255.
img=np.clip(img,0,255)
img=np.round(img,0).astype(np.uint8)
cv2_imsave(join(save_path, '{:0>4}.png'.format(i*num_once+j)),img)
all_time=np.array(all_time)
if max_frame>0:
all_time=np.array(all_time)
print('spent {} s in total and {} s in average'.format(np.sum(all_time),np.mean(all_time[1:])))
def submit_(filename):
if filename.name in TRAIN_FN:
image = cv2_imread(str(filename), "image")
image_post = global_hist_equalizer(image)
cv2.imwrite(f"{args.dst}/{filename.name}", image_post)
else:
shutil.copy(args.src.joinpath(filename.name),
args.dst.joinpath(filename.name))
def eval(self):
print('Evaluating ...')
if not hasattr(self, 'sess'):
sess = tf.Session()
self.load(sess, self.save_dir)
else:
sess = self.sess
border=8
in_h,in_w=self.eval_in_size
out_h = in_h*self.scale #512
out_w = in_w*self.scale #960
bd=border//self.scale
eval_gt = tf.placeholder(tf.float32, [None, self.num_frames, out_h, out_w, 3])
eval_inp=DownSample(eval_gt, BLUR, scale=self.scale)
filenames=open(self.eval_dir, 'rt').read().splitlines()
# sorted(glob.glob(join(self.eval_dir,'*')))
# gt_list=[sorted(glob.glob(join(f,'truth','*.png'))) for f in filenames]
gt_list=[sorted(glob.glob(join('H:/AI4K/data/frame_data/training/HR', f, '*.png'))) for f in filenames]
gt_list = gt_list[:3]
center=15
batch_gt = []
batch_cnt=0
mse_acc=None
for gtlist in gt_list:
max_frame=len(gtlist)
for idx0 in range(center, max_frame, 32):
index=np.array([i for i in range(idx0-self.num_frames//2,idx0+self.num_frames//2+1)])
index=np.clip(index,0,max_frame-1).tolist()
gt=[cv2_imread(gtlist[i]) for i in index]
gt = [i[border:out_h+border, border:out_w+border, :].astype(np.float32) / 255.0 for i in gt]
batch_gt.append(np.stack(gt, axis=0))
if len(batch_gt) == self.eval_basz:
batch_gt = np.stack(batch_gt, 0)
batch_lr=sess.run(eval_inp,feed_dict={eval_gt:batch_gt})
mse_val=sess.run(self.eval_mse,feed_dict={self.L_eval:batch_lr, self.H:batch_gt[:,self.num_frames//2:self.num_frames//2+1]})
if mse_acc is None:
mse_acc = mse_val
else:
mse_acc = np.concatenate([mse_acc, mse_val], axis=0)
batch_gt = []
print('\tEval batch {} - {} ...'.format(batch_cnt, batch_cnt + self.eval_basz))
batch_cnt+=self.eval_basz
psnr_acc = 10 * np.log10(1.0 / mse_acc)
mse_avg = np.mean(mse_acc, axis=0)
psnr_avg = np.mean(psnr_acc, axis=0)
for i in range(mse_avg.shape[0]):
tf.summary.scalar('val_mse{}'.format(i), tf.convert_to_tensor(mse_avg[i], dtype=tf.float32))
print('Eval PSNR: {}, MSE: {}'.format(psnr_avg, mse_avg))
# write to log file
with open(self.log_dir, 'a+') as f:
mse_avg=(mse_avg*1e6).astype(np.int64)/(1e6)
psnr_avg=(psnr_avg*1e6).astype(np.int64)/(1e6)
f.write('{'+'"Iter": {} , "PSNR": {}, "MSE": {}'.format(sess.run(self.global_step), psnr_avg.tolist(), mse_avg.tolist())+'}\n')
def test_video_lr(self, path, name='result', reuse=False, part=8):
save_path = join(path, name)
automkdir(save_path)
inp_path = join(path, 'blur{}'.format(self.scale))
imgs = sorted(glob.glob(join(inp_path, '*.png')))
imgs = np.array([cv2_imread(i) / 255. for i in imgs])
lr_list = []
max_frame = imgs.shape[0]
for i in range(max_frame):
index = np.array([
i for i in range(i - self.num_frames // 2, i +
self.num_frames // 2 + 1)
])
index = np.clip(index, 0, max_frame - 1).tolist()
lr_list.append(np.array([imgs[j] for j in index]))
lr_list = np.array(lr_list)
if not reuse:
self.build()
sess = tf.Session()
self.sess = sess
sess.run(tf.global_variables_initializer())
self.saver = tf.train.Saver(max_to_keep=100,
keep_checkpoint_every_n_hours=1)
self.load(sess, self.save_dir)
print('Save at {}'.format(save_path))
print('{} Inputs With Shape {}'.format(lr_list.shape[0],
lr_list.shape[1:]))
part = min(part, max_frame)
if max_frame % part == 0:
num_once = max_frame // part
else:
num_once = max_frame // part + 1
all_time = 0
for i in trange(part):
st_time = time.time()
sr = self.sess.run(self.SR,
feed_dict={
self.L:
lr_list[i * num_once:(i + 1) * num_once],
self.is_train: False
})
onece_time = time.time() - st_time
if i > 0:
all_time += onece_time
for j in range(sr.shape[0]):
img = sr[j][0] * 255.
img = np.clip(img, 0, 255).astype(np.uint8)
imgname = '{:0>4}.png'.format(i * num_once + j)
cv2_imsave(join(save_path, imgname), img)
print('spent {} s in total and {} s in average'.format(
all_time, all_time / (max_frame - 1)))
def test_video_lr(self, path, output_path, exp_name='PFNL_result'):
num_once = 1
_, video_name = os.path.split(path)
save_path = join(output_path, exp_name, video_name)
automkdir(save_path)
imgs = sorted(glob.glob(join(path, '*.png')))
max_frame = len(imgs)
lrs = np.array([cv2_imread(i) for i in imgs]) / 255.
lrs = lrs.astype('float32')
h, w, _ = lrs[0].shape
lr_list = []
for i in range(max_frame):
index = np.array([
i for i in range(i - self.num_frames // 2, i +
self.num_frames // 2 + 1)
])
index = np.clip(index, 0, max_frame - 1).tolist()
lr_list.append(np.array([lrs[j] for j in index]))
del lrs
lr_list = np.array(lr_list)
# L_test = tf.placeholder(tf.float32, shape=[1, self.num_frames, h, w, 3], name='L_test')
# SR_test=self.forward(L_test)
print('Save at {}'.format(save_path))
print('{} Inputs With Shape {}'.format(max_frame, [h, w]))
all_time = []
for i in trange(max_frame):
st_time = time.time()
sr = self.sess.run(self.SR_test,
feed_dict={
self.L_test:
lr_list[i * num_once:(i + 1) * num_once]
})
all_time.append(time.time() - st_time)
for j in range(sr.shape[0]):
img = sr[j] * 255.
img = np.clip(img, 0, 255)
img = np.round(img, 0).astype(np.uint8)
cv2_imsave(
join(save_path, '{:0>4}.jpg'.format(i * num_once + j)),
img, 100)
all_time = np.array(all_time)
if max_frame > 0:
all_time = np.array(all_time)
print('spent {} s in total and {} s in average'.format(
np.sum(all_time), np.mean(all_time[1:])))
del imgs
# del L_test
# del SR_test
# del lrs
del lr_list
def test_video(self, path, name='result', reuse=False):
save_path = join(path, name)
automkdir(save_path)
inp_path = join(path, 'blur{}'.format(self.scale))
imgs = sorted(glob.glob(join(inp_path, '*.png')))
imgs = np.array([cv2_imread(i) / 255. for i in imgs])
n, h, w, c = imgs.shape
max_frame = n
self.L = tf.placeholder(tf.float32, shape=[1, h, w, 3], name='L_input')
self.LP = tf.placeholder(tf.float32,
shape=[1, h, w, 3],
name='Previous_L_input')
self.est = tf.placeholder(tf.float32,
shape=[1, h * self.scale, w * self.scale, 3],
name='est')
self.sr0 = self.forward(self.L)
self.sr1 = self.forward(self.L, self.LP, self.est)
if not reuse:
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
#sess=tf.Session()
self.sess = sess
sess.run(tf.global_variables_initializer())
self.saver = tf.train.Saver(max_to_keep=100,
keep_checkpoint_every_n_hours=1)
self.load(sess, self.save_dir)
print('Save at {}'.format(save_path))
print('{} Inputs With Shape {}'.format(imgs.shape[0], imgs.shape[1:]))
all_time = []
for i in trange(max_frame):
st_time = time.time()
if i == 0:
SR = self.sess.run(self.sr0, feed_dict={self.L: imgs[i:i + 1]})
else:
SR = self.sess.run(self.sr1,
feed_dict={
self.L: imgs[i:i + 1],
self.LP: imgs[i - 1:i],
self.est: SR
})
all_time.append(time.time() - st_time)
img = SR[0] * 255.
img = np.clip(img, 0, 255).astype(np.uint8)
cv2_imsave(join(save_path, '{:0>4}.png'.format(i)), img)
if max_frame > 0:
all_time = np.array(all_time)
print('spent {} s in total and {} s in average'.format(
np.sum(all_time), np.mean(all_time[1:])))
def eval(self):
print('Evaluating ...')
if not hasattr(self, 'sess'):
global_step = tf.Variable(initial_value=0, trainable=False)
self.global_step = global_step
self.build()
sess = tf.Session()
self.load(sess, self.save_dir)
else:
sess = self.sess
border = 8
in_h, in_w = self.eval_in_size
out_h = in_h * self.scale #512
out_w = in_w * self.scale #960
bd = border // self.scale
filenames = open(self.eval_dir, 'rt').read().splitlines()
hr_list = [
sorted(glob.glob(join(f, 'truth', '*.png'))) for f in filenames
]
lr_list = [
sorted(glob.glob(join(f, 'blur{}'.format(self.scale), '*.png')))
for f in filenames
]
center = 15
batch_hr = []
batch_lr = []
batch_cnt = 0
mse_acc = None
for lrlist, hrlist in zip(lr_list, hr_list):
max_frame = len(lrlist)
for idx0 in range(center, max_frame, 32):
index = np.array([
i for i in range(idx0 - self.num_frames // 2, idx0 +
self.num_frames // 2 + 1)
])
index = np.clip(index, 0, max_frame - 1).tolist()
gt = [cv2_imread(hrlist[idx0])]
inp = [cv2_imread(lrlist[i]) for i in index]
inp = [
i[bd:in_h + bd, bd:in_w + bd].astype(np.float32) / 255.0
for i in inp
]
gt = [
i[border:out_h + border, border:out_w + border, :].astype(
np.float32) / 255.0 for i in gt
]
batch_hr.append(np.stack(gt, axis=0))
batch_lr.append(np.stack(inp, axis=0))
if len(batch_hr) == self.eval_basz:
batch_hr = np.stack(batch_hr, 0)
batch_lr = np.stack(batch_lr, 0)
mse_val = sess.run(self.eval_mse,
feed_dict={
self.L: batch_lr,
self.H: batch_hr,
self.is_train: False
})
if mse_acc is None:
mse_acc = mse_val
else:
mse_acc = np.concatenate([mse_acc, mse_val], axis=0)
batch_hr = []
batch_lr = []
print('\tEval batch {} - {} ...'.format(
batch_cnt, batch_cnt + self.eval_basz))
batch_cnt += self.eval_basz
psnr_acc = 10 * np.log10(1.0 / mse_acc)
mse_avg = np.mean(mse_acc, axis=0)
psnr_avg = np.mean(psnr_acc, axis=0)
for i in range(mse_avg.shape[0]):
tf.summary.scalar(
'val_mse{}'.format(i),
tf.convert_to_tensor(mse_avg[i], dtype=tf.float32))
print('Eval PSNR: {}, MSE: {}'.format(psnr_avg, mse_avg))
# write to log file
with open(self.log_dir, 'a+') as f:
mse_avg = (mse_avg * 1e6).astype(np.int64) / (1e6)
psnr_avg = (psnr_avg * 1e6).astype(np.int64) / (1e6)
f.write('{' + '"Iter": {} , "PSNR": {}, "MSE": {}'.format(
sess.run(self.global_step), psnr_avg.tolist(),
mse_avg.tolist()) + '}\n')
def testvideo(self,
dataPath=None,
savename='result',
reuse=False,
scale=4,
num_frames=3):
scale = self.scale
num_frames = self.num_frames
inList = sorted(
glob.glob(os.path.join(dataPath, 'blur{}/*.png').format(scale)))
inp = [cv2_imread(i).astype(np.float32) / 255.0 for i in inList]
max_frame = len(inList)
print('Testing path: {}'.format(dataPath))
print('# of testing frames: {}'.format(len(inList)))
savepath = os.path.join(dataPath, savename)
automkdir(savepath)
all_time = []
for idx0 in trange(len(inList)):
T = num_frames // 2
imgs = [inp[0] for i in np.arange(idx0 - T, 0)]
imgs.extend([inp[i] for i in np.arange(max(0, idx0 - T), idx0)])
imgs.extend([
inp[i] for i in np.arange(idx0, min(len(inList), idx0 + T + 1))
])
imgs.extend(
[inp[-1] for i in np.arange(idx0 + T,
len(inList) - 1, -1)])
dims = imgs[0].shape
if len(dims) == 2:
imgs = [np.expand_dims(i, -1) for i in imgs]
h, w, c = imgs[0].shape
out_h = h * scale
out_w = w * scale
padh = int(ceil(h / 4.0) * 4.0 - h)
padw = int(ceil(w / 4.0) * 4.0 - w)
imgs = [
np.pad(i, [[0, padh], [0, padw], [0, 0]], 'edge') for i in imgs
]
imgs = np.expand_dims(np.stack(imgs, axis=0), 0)
if idx0 == 0:
# frames_lr = tf.convert_to_tensor(imgs, tf.float32)
frames_lr = tf.placeholder(dtype=tf.float32, shape=imgs.shape)
frames_ref_ycbcr = rgb2ycbcr(frames_lr[:, T:T + 1, :, :, :])
# frames_ref_ycbcr = tf.tile(frames_ref_ycbcr, [1, num_frames, 1, 1, 1])
output = self.forward(frames_lr,
is_training=False,
reuse=reuse)
# print (frames_lr_ycbcr.get_shape(), h, w, padh, padw)
if len(dims) == 3:
output_rgb = ycbcr2rgb(
tf.concat([
output,
resize_images(
frames_ref_ycbcr, [(h + padh) * scale,
(w + padw) * scale],
method=2)[:, :, :, :, 1:3]
], -1))
else:
output_rgb = output
output = output[:, :, :out_h, :out_w, :]
output_rgb = output_rgb[:, :, :out_h, :out_w, :]
if reuse == False:
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
#sess=tf.Session()
self.sess = sess
self.saver = tf.train.Saver(max_to_keep=50,
keep_checkpoint_every_n_hours=1)
self.load(self.sess, self.save_dir)
reuse = True
st_time = time.time()
[imgs_hr, imgs_hr_rgb,
uv] = self.sess.run([output, output_rgb, self.uv],
feed_dict={frames_lr: imgs})
all_time.append(time.time() - st_time)
if len(dims) == 3:
cv2_imsave(os.path.join(savepath, 'rgb_%03d.png' % (idx0)),
im2uint8(imgs_hr_rgb[0, -1, :, :, :]))
print('SR results path: {}'.format(savepath))
if max_frame > 0:
all_time = np.array(all_time)
print('spent {} s in total and {} s in average'.format(
np.sum(all_time), np.mean(all_time[1:])))
def evaluation(self):
print('Evaluating ...')
filenames = open(self.eval_dir, 'rt').read().splitlines()
gtList_all = [
sorted(glob.glob(join(f, 'truth', '*.png'))) for f in filenames
]
inList_all = [
sorted(glob.glob(join(f, 'blur{}'.format(self.scale), '*.png')))
for f in filenames
]
if not hasattr(self, 'sess'):
sess = tf.Session()
else:
sess = self.sess
border = 8
in_h, in_w = self.eval_in_size
out_h = in_h * self.scale #512
out_w = in_w * self.scale #960
bd = border // self.scale
if not hasattr(self, 'eval_input'):
self.eval_input = tf.placeholder(
tf.float32,
[self.eval_batch_size, self.num_frames, in_h, in_w, 3])
self.eval_gt = tf.placeholder(
tf.float32, [self.eval_batch_size, 1, out_h, out_w, 3])
self.eval_output = self.forward(self.eval_input,
is_training=False,
reuse=True)
# calculate loss
frame_gt_y = rgb2y(self.eval_gt)
self.eval_mse = tf.reduce_mean(
(self.eval_output[:, :, :, :, :] - frame_gt_y)**2,
axis=[2, 3, 4])
batch_in = []
batch_gt = []
radius = self.num_frames // 2
mse_acc = None
ssim_acc = None
batch_cnt = 0
for inList, gtList in zip(inList_all, gtList_all):
for idx0 in range(15, len(inList), 32):
inp = [cv2_imread(inList[0]) for i in range(idx0 - radius, 0)]
inp.extend([
cv2_imread(inList[i])
for i in range(max(0, idx0 - radius), idx0)
])
inp.extend([
cv2_imread(inList[i])
for i in range(idx0, min(len(inList), idx0 + radius + 1))
])
inp.extend([
cv2_imread(inList[-1])
for i in range(idx0 + radius,
len(inList) - 1, -1)
])
inp = [
i[bd:in_h + bd, bd:in_w + bd, :].astype(np.float32) / 255.0
for i in inp
]
gt = [cv2_imread(gtList[idx0])]
gt = [
i[border:out_h + border, border:out_w + border, :].astype(
np.float32) / 255.0 for i in gt
]
batch_in.append(np.stack(inp, axis=0))
batch_gt.append(np.stack(gt, axis=0))
if len(batch_in) == self.eval_batch_size:
batch_cnt += self.eval_batch_size
batch_in = np.stack(batch_in, 0)
batch_gt = np.stack(batch_gt, 0)
mse_val, eval_output_val = sess.run(
[self.eval_mse, self.eval_output],
feed_dict={
self.eval_input: batch_in,
self.eval_gt: batch_gt
})
ssim_val = np.array([[
compute_ssim(eval_output_val[ib, it, :, :, 0],
batch_gt[ib, 0, :, :, 0],
l=1.0) for it in range(1)
] for ib in range(self.eval_batch_size)])
if mse_acc is None:
mse_acc = mse_val
ssim_acc = ssim_val
else:
mse_acc = np.concatenate([mse_acc, mse_val], axis=0)
ssim_acc = np.concatenate([ssim_acc, ssim_val], axis=0)
batch_in = []
batch_gt = []
print('\tEval batch {} - {} ...'.format(
batch_cnt, batch_cnt + self.eval_batch_size))
psnr_acc = 10 * np.log10(1.0 / mse_acc)
mse_avg = np.mean(mse_acc, axis=0)
psnr_avg = np.mean(psnr_acc, axis=0)
ssim_avg = np.mean(ssim_acc, axis=0)
for i in range(mse_avg.shape[0]):
tf.summary.scalar(
'val_mse{}'.format(i),
tf.convert_to_tensor(mse_avg[i], dtype=tf.float32))
print('Eval MSE: {}, PSNR: {}'.format(mse_avg, psnr_avg))
# write to log file
with open(self.log_dir, 'a+') as f:
mse_avg = (mse_avg * 1e8).astype(np.int64) / (1e8)
psnr_avg = (psnr_avg * 1e8).astype(np.int64) / (1e8)
ssim_avg = (ssim_avg * 1e8).astype(np.int64) / (1e8)
f.write('{' +
'"Iter": {} , "MSE": {}, "PSNR": {}, "SSIM": {}'.format(
sess.run(self.global_step), mse_avg.tolist(),
psnr_avg.tolist(), ssim_avg.tolist()) + '}\n')
def eval(self, epoch, loss_epoch):
print('Evaluating on the validation set...')
psnr_all = 0
ssim_all = 0
count = 0
for video_index in range(0, len(self.eval_frame_data_LR)):
cur_video_LR = self.eval_frame_data_LR[video_index]
path, _ = os.path.split(cur_video_LR[0])
_, name = os.path.split(path)
cur_video_HR = self.eval_frame_data_HR[video_index]
max_frame = len(cur_video_LR)
# temp_img = cv2_imread(cur_video_LR[0])
# h,w,_ = temp_img.shape
# L_eval = tf.placeholder(tf.float32, shape=[1, self.num_frames, h, w, 3], name='L_test')
# SR_test = self.forward(L_eval)
for i in range(max_frame):
start_time = time.time()
count += 1
index = np.array([
i for i in range(i - self.num_frames // 2, i +
self.num_frames // 2 + 1)
])
index = np.clip(index, 0, max_frame - 1).tolist()
lrs = np.array([cv2_imread(cur_video_LR[i])
for i in index]) / 255.0
lrs = lrs.astype('float32')
lrs = np.expand_dims(lrs, 0)
sr = self.sess.run(self.SR_test, feed_dict={self.L_test: lrs})
sr = sr * 255
sr = np.squeeze(sr, axis=(0))
sr = np.clip(sr, 0, 255)
sr = np.round(sr, 0).astype(np.uint8)
hr = cv2_imread(cur_video_HR[i])
psnr = compare_psnr(sr, hr)
ssim = compare_ssim(sr, hr, multichannel=True)
psnr_all += psnr
ssim_all += ssim
cost_time = time.time() - start_time
print(
'[*Epoch:{:05d}] val video:{} -> frame:{:05d}, psnr={:.4f}, ssim={:.4f}, cost time={:.2f} '
.format(epoch, name, i, psnr, ssim, cost_time))
a_psnr = psnr_all / count
a_ssim = ssim_all / count
eval_ss = self.sess.run(self.merge_op_eval,
feed_dict={
self.loss_epoch: loss_epoch,
self.psnr_eval: a_psnr,
self.ssim_eval: a_ssim
})
self.writer.add_summary(eval_ss, epoch)
print(
'{' +
'"Epoch": {:05d} , "Training Loss":{:.6f}, " Eval PSNR": {:.4f}, "Eval SSIM": {:.4f}'
.format(epoch, loss_epoch, a_psnr, a_ssim) + '}')
# write to log file
with open(self.log_dir, 'a+') as f:
f.write(
'{' +
'"Epoch": {:05d} , "Training Loss":{:.6f}, " Eval PSNR": {:.4f}, "Eval SSIM": {:.4f}'
.format(epoch, loss_epoch, a_psnr, a_ssim) + '}\n')
def testvideo(self,
dataPath=None,
savename='result',
reuse=False,
scale_factor=4,
num_frames=3):
inList = sorted(
glob.glob(
os.path.join(dataPath, 'blur{}/*.png').format(scale_factor)))
inp = [cv2_imread(i).astype(np.float32) / 255.0 for i in inList]
print('Testing path: {}'.format(dataPath))
print('# of testing frames: {}'.format(len(inList)))
max_frame = len(inList)
savepath = os.path.join(dataPath, savename)
if not os.path.exists(savepath):
os.mkdir(savepath)
self.scale_factor = scale_factor
all_time = []
for idx0 in trange(len(inList)):
T = num_frames // 2
imgs = [inp[0] for i in np.arange(idx0 - T, 0)]
imgs.extend([inp[i] for i in np.arange(max(0, idx0 - T), idx0)])
imgs.extend([
inp[i] for i in np.arange(idx0, min(len(inList), idx0 + T + 1))
])
imgs.extend(
[inp[-1] for i in np.arange(idx0 + T,
len(inList) - 1, -1)])
dims = imgs[0].shape
if len(dims) == 2:
imgs = [np.expand_dims(i, -1) for i in imgs]
h, w, c = imgs[0].shape
out_h = h * scale_factor
out_w = w * scale_factor
padh = int(ceil(h / 4.0) * 4.0 - h)
padw = int(ceil(w / 4.0) * 4.0 - w)
imgs = [
np.pad(i, [[0, padh], [0, padw], [0, 0]], 'edge') for i in imgs
]
imgs = np.expand_dims(np.stack(imgs, axis=0), 0)
if idx0 == 0:
# frames_lr = tf.convert_to_tensor(imgs, tf.float32)
frames_lr = tf.placeholder(dtype=tf.float32, shape=imgs.shape)
frames_ref_ycbcr = rgb2ycbcr(frames_lr[:, T:T + 1, :, :, :])
frames_ref_ycbcr = tf.tile(frames_ref_ycbcr,
[1, num_frames, 1, 1, 1])
output, frame_i_fw = self.forward(frames_lr,
is_training=False,
reuse=reuse)
# print (frames_lr_ycbcr.get_shape(), h, w, padh, padw)
if len(dims) == 3:
output_rgb = ycbcr2rgb(
tf.concat([
output,
resize_images(
frames_ref_ycbcr, [(h + padh) * scale_factor,
(w + padw) * scale_factor],
method=2)[:, :, :, :, 1:3]
], -1))
else:
output_rgb = output
output = output[:, :, :out_h, :out_w, :]
output_rgb = output_rgb[:, :, :out_h, :out_w, :]
if reuse == False:
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
#sess=tf.Session()
self.sess = sess
# summary_op = tf.summary.merge_all()
# summary_writer = tf.summary.FileWriter('E:/PyProject/out0', sess.graph, flush_secs=30)
self.saver = tf.train.Saver(max_to_keep=50,
keep_checkpoint_every_n_hours=1)
self.load(self.sess, self.save_dir)
#self.flownets.load_easyflow(self.sess, os.path.join('./easyflow_log/model1', 'checkpoints'))
reuse = True
# frames_ref_warp = imwarp_backward(self.uv, tf.tile(tf.expand_dims(self.frame_ref_y, 1), [1, num_frames, 1, 1, 1]),
# [200, 244])
# warp_err = tf.reduce_mean((self.frames_y - frames_ref_warp) ** 2, axis=4)
case_path = dataPath.split('/')[-1]
# print('Testing - ', case_path, len(imgs))
st_time = time.time()
[imgs_hr, imgs_hr_rgb, uv, frame_fw
] = self.sess.run([output, output_rgb, self.uv, frame_i_fw],
feed_dict={frames_lr: imgs})
all_time.append(time.time() - st_time)
# print('spent {} s'.format(ed_time-st_time))
'''cv2_imsave(os.path.join(savepath, 'y_%03d.png' % (idx0)),
im2uint8(imgs_hr[0, -1, :, :, 0]))'''
'''for q in range(0, 3):
cv2_imsave(os.path.join(savepath, 'y_ff%03d_%01d.png' % (idx0, q)),
im2uint8(frame_fw[q, :, :, 0]))'''
if len(dims) == 3:
cv2_imsave(os.path.join(savepath, '%04d.png' % (idx0)),
im2uint8(imgs_hr_rgb[0, -1, :, :, :]))
# summary_str = self.sess.run(summary_op)
# summary_writer.add_summary(summary_str, idx0)
print('SR results path: {}'.format(savepath))
if max_frame > 0:
all_time = np.array(all_time)
print('spent {} s in total and {} s in average'.format(
np.sum(all_time), np.mean(all_time[1:])))
def test_video_lr(self, path, output_path, name='PFNL_result', reuse=False, part=10000):
_, video_name = os.path.split(path)
save_path=join(output_path, video_name, name)
automkdir(save_path)
# inp_path=join(path,'blur{}'.format(self.scale))
imgs=sorted(glob.glob(join(path,'*.png')))
# names = os.listdir(path)
# imgs = []
# for name_ in names:
# if name_[-3]
# temp_path = os.path.join(path, name_)
# imgs.append(temp_path)
max_frame=len(imgs)
# lrs=np.array([cv2_imread(i) for i in imgs])/255.
lrs = []
for i in imgs:
img = cv2_imread(i)
img = cv2.resize(img, (int(img.shape[1]*0.5), int(img.shape[0]*0.5)))
img = img / 255
lrs.append(img)
lrs = np.array(lrs)
if part>max_frame:
part=max_frame
if max_frame%part ==0 :
num_once=max_frame//part
else:
num_once=max_frame//part+1
h,w,c=lrs[0].shape
L_test = tf.placeholder(tf.float32, shape=[num_once, self.num_frames, h, w, 3], name='L_test')
SR_test=self.forward(L_test)
if not reuse:
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
#sess=tf.Session()
self.sess=sess
sess.run(tf.global_variables_initializer())
self.saver = tf.train.Saver(max_to_keep=100, keep_checkpoint_every_n_hours=1)
self.load(sess, self.save_dir)
lr_list=[]
max_frame=lrs.shape[0]
for i in range(max_frame):
index=np.array([i for i in range(i-self.num_frames//2,i+self.num_frames//2+1)])
index=np.clip(index,0,max_frame-1).tolist()
lr_list.append(np.array([lrs[j] for j in index]))
lr_list=np.array(lr_list)
print('Save at {}'.format(save_path))
print('{} Inputs With Shape {}'.format(lrs.shape[0],lrs.shape[1:]))
h,w,c=lrs.shape[1:]
all_time=[]
for i in trange(part):
st_time=time.time()
sr=self.sess.run(SR_test,feed_dict={L_test : lr_list[i*num_once:(i+1)*num_once]})
all_time.append(time.time()-st_time)
for j in range(sr.shape[0]):
img=sr[j][0]*255.
img=np.clip(img,0,255)
img=np.round(img,0).astype(np.uint8)
cv2_imsave(join(save_path, '{:0>4}.png'.format(i*num_once+j)),img)
all_time=np.array(all_time)
if max_frame>0:
all_time=np.array(all_time)
print('spent {} s in total and {} s in average'.format(np.sum(all_time),np.mean(all_time[1:])))
def makeup(self,
path,
output_path,
exp_name='PFNL_result',
method='replicate'):
print(' [**] Using makeup method: ' + method)
_, video_name = os.path.split(path)
save_path = join(output_path, exp_name, method, video_name)
automkdir(save_path)
imgs = sorted(glob.glob(join(path, '*.png')))
max_frame = len(imgs)
lrs = np.array([cv2_imread(i) for i in imgs]) / 255.
lrs = lrs.astype('float32')
h, w, _ = lrs[0].shape
lr_list = []
deal_index_1 = np.linspace(0,
self.num_frames // 2 - 1,
self.num_frames // 2,
dtype=np.int)
deal_index_2 = np.linspace(max_frame - self.num_frames // 2,
max_frame - 1,
self.num_frames // 2,
dtype=np.int)
deal_index = np.concatenate((deal_index_1, deal_index_2), axis=0)
# max_frame = len(deal_index)
for i in deal_index:
index = np.array([
i for i in range(i - self.num_frames // 2, i +
self.num_frames // 2 + 1)
])
if method == 'replicate':
# index=np.array([i for i in range(i-self.num_frames//2,i+self.num_frames//2+1)])
index = np.clip(index, 0, max_frame - 1)
elif method == 'reflection':
# index=np.array([i for i in range(i-self.num_frames//2,i+self.num_frames//2+1)])
for ii in range(len(index)):
index[ii] = index[ii] if index[ii] >= 0 else index[ii] * -1
index[ii] = index[ii] if index[ii] < max_frame else 2 * (
max_frame - 1) - index[ii]
elif method == 'new_info':
# index=np.array([i for i in range(i-self.num_frames//2,i+self.num_frames//2+1)])
for ii in range(len(index)):
index[ii] = index[
ii] if index[ii] >= 0 else index[ii] * -1 + index[-1]
index[ii] = index[ii] if index[ii] < max_frame else index[
0] - (index[ii] - max_frame + 1)
elif method == 'circle':
for ii in range(len(index)):
index[ii] = index[
ii] if index[ii] >= 0 else self.num_frames + index[ii]
index[ii] = index[ii] if index[ii] < max_frame else index[
ii] - i + index[0] - self.num_frames // 2 - 1
index = index.tolist()
lr_list.append(np.array([lrs[j] for j in index]))
del lrs
lr_list = np.array(lr_list)
L_test = tf.placeholder(tf.float32,
shape=[1, self.num_frames, h, w, 3],
name='L_test')
SR_test = self.forward(L_test)
print('Save at {}'.format(save_path))
print('{} Inputs With Shape {}'.format(max_frame, [h, w]))
all_time = []
for i in trange(len(deal_index)):
# i = deal_index[cc]
st_time = time.time()
sr = self.sess.run(SR_test, feed_dict={L_test: lr_list[i:(i + 1)]})
all_time.append(time.time() - st_time)
for j in range(sr.shape[0]):
img = sr[j][0] * 255.
img = np.clip(img, 0, 255)
img = np.round(img, 0).astype(np.uint8)
cv2_imsave(join(save_path, '{:0>4}.png'.format(deal_index[i])),
img, 100)
all_time = np.array(all_time)
if max_frame > 0:
all_time = np.array(all_time)
print('spent {} s in total and {} s in average'.format(
np.sum(all_time), np.mean(all_time[1:])))
del imgs
del L_test
del SR_test
# del lrs
del lr_list
