In [ ]:
import fastai
from fastai.vision import *
from fastai.callbacks import *
from fastai.vision.gan import *
In [ ]:
path = untar_data(URLs.PETS)
path_hr = path/'images'
path_lr = path/'crappy'
Prepare the input data by crappifying images.
使用噪化的图片准备输入数据
In [ ]:
from crappify import *
Uncomment the first time you run this notebook.
如果首次运行这个notebook，请把注释打开
In [ ]:
#il = ImageList.from_folder(path_hr)
#parallel(crappifier(path_lr, path_hr), il.items)
For gradual resizing we can change the commented line here.
如果需要对图片进行渐变修改或者缩放尺寸请打开下面的注释
In [ ]:
bs,size=32, 128
# bs,size = 24,160
#bs,size = 8,256
arch = models.resnet34
Now let's pretrain the generator.
现在我们来预训练数据生成器
In [ ]:
arch = models.resnet34
src = ImageImageList.from_folder(path_lr).split_by_rand_pct(0.1, seed=42)
In [ ]:
def get_data(bs,size):
data = (src.label_from_func(lambda x: path_hr/x.name)
.transform(get_transforms(max_zoom=2.), size=size, tfm_y=True)
.databunch(bs=bs).normalize(imagenet_stats, do_y=True))
data.c = 3
return data
In [ ]:
data_gen = get_data(bs,size)
In [ ]:
data_gen.show_batch(4)
In [ ]:
wd = 1e-3
In [ ]:
y_range = (-3.,3.)
In [ ]:
loss_gen = MSELossFlat()
In [ ]:
def create_gen_learner():
return unet_learner(data_gen, arch, wd=wd, blur=True, norm_type=NormType.Weight,
self_attention=True, y_range=y_range, loss_func=loss_gen)
In [ ]:
learn_gen = create_gen_learner()
In [ ]:
learn_gen.fit_one_cycle(2, pct_start=0.8)
In [ ]:
learn_gen.unfreeze()
In [ ]:
learn_gen.fit_one_cycle(3, slice(1e-6,1e-3))
In [ ]:
learn_gen.show_results(rows=4)
In [ ]:
learn_gen.save('gen-pre2')
In [ ]:
learn_gen.load('gen-pre2');
In [ ]:
name_gen = 'image_gen'
path_gen = path/name_gen
In [ ]:
# shutil.rmtree(path_gen)
In [ ]:
path_gen.mkdir(exist_ok=True)
In [ ]:
def save_preds(dl):
i=0
names = dl.dataset.items
for b in dl:
preds = learn_gen.pred_batch(batch=b, reconstruct=True)
for o in preds:
o.save(path_gen/names[i].name)
i += 1
In [ ]:
save_preds(data_gen.fix_dl)
In [ ]:
PIL.Image.open(path_gen.ls()[0])
Out[ ]:
In [ ]:
learn_gen=None
gc.collect()
Out[ ]:
Pretrain the critic on crappy vs not crappy.
使用噪化和非噪化数据来预训练批评者网络
In [ ]:
def get_crit_data(classes, bs, size):
src = ImageList.from_folder(path, include=classes).split_by_rand_pct(0.1, seed=42)
ll = src.label_from_folder(classes=classes)
data = (ll.transform(get_transforms(max_zoom=2.), size=size)
.databunch(bs=bs).normalize(imagenet_stats))
data.c = 3
return data
In [ ]:
data_crit = get_crit_data([name_gen, 'images'], bs=bs, size=size)
In [ ]:
data_crit.show_batch(rows=3, ds_type=DatasetType.Train, imgsize=3)
In [ ]:
loss_critic = AdaptiveLoss(nn.BCEWithLogitsLoss())
In [ ]:
def create_critic_learner(data, metrics):
return Learner(data, gan_critic(), metrics=metrics, loss_func=loss_critic, wd=wd)
In [ ]:
learn_critic = create_critic_learner(data_crit, accuracy_thresh_expand)
In [ ]:
learn_critic.fit_one_cycle(6, 1e-3)
In [ ]:
learn_critic.save('critic-pre2')
Now we'll combine those pretrained model in a GAN.
现在我们将这些经过预训练的模型整合到GAN里。
In [ ]:
learn_crit=None
learn_gen=None
gc.collect()
Out[ ]:
In [ ]:
data_crit = get_crit_data(['crappy', 'images'], bs=bs, size=size)
In [ ]:
learn_crit = create_critic_learner(data_crit, metrics=None).load('critic-pre2')
In [ ]:
learn_gen = create_gen_learner().load('gen-pre2')
To define a GAN Learner, we just have to specify the learner objects foor the generator and the critic. The switcher is a callback that decides when to switch from discriminator to generator and vice versa. Here we do as many iterations of the discriminator as needed to get its loss back < 0.5 then one iteration of the generator.
为了定义一个GAN学习器，我们需要为生成器和评价网络指定学习组件对象。switcher（切换器）是一个回调（装置）,这个装置决定何时在鉴别器和生成器之间转换。这里我们针对鉴别器训练足够多的iterations（迭代次数）直到其损失函数值小于0.5，然后再对生成器训练一次。
The loss of the critic is given by learn_crit.loss_func. We take the average of this loss function on the batch of real predictions (target 1) and the batch of fake predicitions (target 0).
评价网络的损失由learn_crit.loss_func获得。基于正预测值(目标为1)和负预测值(目标为0)的迭代训练会针对这个损失函数取均值。
The loss of the generator is weighted sum (weights in weights_gen) of learn_crit.loss_func on the batch of fake (passed throught the critic to become predictions) with a target of 1, and the learn_gen.loss_func applied to the output (batch of fake) and the target (corresponding batch of superres images).
生成器网络的损失则是learn_crit.loss_func和learn_gen.loss_func的加权和（权重在weights_gen中），（其中）learn_crit.loss_func是基于目标为1的fake（失败的）批次（通过评价网络成为预测值），learn_gen.loss_func则应用到fake批次的输出及目标上（对应一批超分辨率图像）。
In [ ]:
switcher = partial(AdaptiveGANSwitcher, critic_thresh=0.65)
learn = GANLearner.from_learners(learn_gen, learn_crit, weights_gen=(1.,50.), show_img=False, switcher=switcher,
opt_func=partial(optim.Adam, betas=(0.,0.99)), wd=wd)
learn.callback_fns.append(partial(GANDiscriminativeLR, mult_lr=5.))
In [ ]:
lr = 1e-4
In [ ]:
learn.fit(40,lr)
In [ ]:
learn.save('gan-1c')
In [ ]:
learn.data=get_data(16,192)
In [ ]:
learn.fit(10,lr/2)
In [ ]:
learn.show_results(rows=16)
In [ ]:
learn.save('gan-1c')