以下链接是个人关于stylegan所有见解，如有错误欢迎大家指出，我会第一时间纠正，如有兴趣可以加微信：17575010159 相互讨论技术。若是帮助到了你什么，一定要记得点赞奥！因为这是对我最大的鼓励。 风格迁移0-00：stylegan-目录-史上最全：https://blog.csdn.net/weixin_43013761/article/details/100895333
源码注释在贴出源码注释之前，我想把Discriminator的结构贴出来把，如果大家已经跑过该程序，应该看到过很多次打印了:
D Params OutputShape WeightShape
--- --- --- ---
images_in - (?, 3, 1024, 1024) -
labels_in - (?, 0) -
lod - () -
FromRGB_lod0 64 (?, 16, 1024, 1024) (1, 1, 3, 16)
1024x1024/Conv0 2320 (?, 16, 1024, 1024) (3, 3, 16, 16)
1024x1024/Conv1_down 4640 (?, 32, 512, 512) (3, 3, 16, 32)
Downscale2D - (?, 3, 512, 512) -
FromRGB_lod1 128 (?, 32, 512, 512) (1, 1, 3, 32)
Grow_lod0 - (?, 32, 512, 512) -
512x512/Conv0 9248 (?, 32, 512, 512) (3, 3, 32, 32)
512x512/Conv1_down 18496 (?, 64, 256, 256) (3, 3, 32, 64)
Downscale2D_1 - (?, 3, 256, 256) -
FromRGB_lod2 256 (?, 64, 256, 256) (1, 1, 3, 64)
Grow_lod1 - (?, 64, 256, 256) -
256x256/Conv0 36928 (?, 64, 256, 256) (3, 3, 64, 64)
256x256/Conv1_down 73856 (?, 128, 128, 128) (3, 3, 64, 128)
Downscale2D_2 - (?, 3, 128, 128) -
FromRGB_lod3 512 (?, 128, 128, 128) (1, 1, 3, 128)
Grow_lod2 - (?, 128, 128, 128) -
128x128/Conv0 147584 (?, 128, 128, 128) (3, 3, 128, 128)
128x128/Conv1_down 295168 (?, 256, 64, 64) (3, 3, 128, 256)
Downscale2D_3 - (?, 3, 64, 64) -
FromRGB_lod4 1024 (?, 256, 64, 64) (1, 1, 3, 256)
Grow_lod3 - (?, 256, 64, 64) -
64x64/Conv0 590080 (?, 256, 64, 64) (3, 3, 256, 256)
64x64/Conv1_down 1180160 (?, 512, 32, 32) (3, 3, 256, 512)
Downscale2D_4 - (?, 3, 32, 32) -
FromRGB_lod5 2048 (?, 512, 32, 32) (1, 1, 3, 512)
Grow_lod4 - (?, 512, 32, 32) -
32x32/Conv0 2359808 (?, 512, 32, 32) (3, 3, 512, 512)
32x32/Conv1_down 2359808 (?, 512, 16, 16) (3, 3, 512, 512)
Downscale2D_5 - (?, 3, 16, 16) -
FromRGB_lod6 2048 (?, 512, 16, 16) (1, 1, 3, 512)
Grow_lod5 - (?, 512, 16, 16) -
16x16/Conv0 2359808 (?, 512, 16, 16) (3, 3, 512, 512)
16x16/Conv1_down 2359808 (?, 512, 8, 8) (3, 3, 512, 512)
Downscale2D_6 - (?, 3, 8, 8) -
FromRGB_lod7 2048 (?, 512, 8, 8) (1, 1, 3, 512)
Grow_lod6 - (?, 512, 8, 8) -
8x8/Conv0 2359808 (?, 512, 8, 8) (3, 3, 512, 512)
8x8/Conv1_down 2359808 (?, 512, 4, 4) (3, 3, 512, 512)
Downscale2D_7 - (?, 3, 4, 4) -
FromRGB_lod8 2048 (?, 512, 4, 4) (1, 1, 3, 512)
Grow_lod7 - (?, 512, 4, 4) -
4x4/MinibatchStddev - (?, 513, 4, 4) -
4x4/Conv 2364416 (?, 512, 4, 4) (3, 3, 513, 512)
4x4/Dense0 4194816 (?, 512) (8192, 512)
4x4/Dense1 513 (?, 1) (512, 1)
scores_out - (?, 1) -
--- --- --- ---
Total 23087249
下面是源码的注释：
#----------------------------------------------------------------------------
# Discriminator used in the StyleGAN paper.
def D_basic(
images_in, # First input: Images [minibatch, channel, height, width].
labels_in, # Second input: Labels [minibatch, label_size].
num_channels = 1, # Number of input color channels. Overridden based on dataset.
resolution = 32, # Input resolution. Overridden based on dataset.
label_size = 0, # Dimensionality of the labels, 0 if no labels. Overridden based on dataset.
fmap_base = 8192, # Overall multiplier for the number of feature maps.
fmap_decay = 1.0, # log2 feature map reduction when doubling the resolution.
fmap_max = 512, # Maximum number of feature maps in any layer.
nonlinearity = 'lrelu', # Activation function: 'relu', 'lrelu',
use_wscale = True, # Enable equalized learning rate?
mbstd_group_size = 4, # Group size for the minibatch standard deviation layer, 0 = disable.
mbstd_num_features = 1, # Number of features for the minibatch standard deviation layer.
dtype = 'float32', # Data type to use for activations and outputs.
fused_scale = 'auto', # True = fused convolution + scaling, False = separate ops, 'auto' = decide automatically.
blur_filter = [1,2,1], # Low-pass filter to apply when resampling activations. None = no filtering.
structure = 'auto', # 'fixed' = no progressive growing, 'linear' = human-readable, 'recursive' = efficient, 'auto' = select automatically.
is_template_graph = False, # True = template graph constructed by the Network class, False = actual evaluation.
**_kwargs): # Ignore unrecognized keyword args.
# 在我们的网络中，输入为(?, 3, 1024,1024),即得到的resolution_log2为10
resolution_log2 = int(np.log2(resolution))
assert resolution == 2**resolution_log2 and resolution >= 4
# 通过stage,stage为fmap网络全链接的层数,stage指定的层数不同，求的的fmap不同,
def nf(stage): return min(int(fmap_base / (2.0 ** (stage * fmap_decay))), fmap_max)
# 进行模糊操作
def blur(x): return blur2d(x, blur_filter) if blur_filter else x
if structure == 'auto': structure = 'linear' if is_template_graph else 'recursive'
act, gain = {'relu': (tf.nn.relu, np.sqrt(2)), 'lrelu': (leaky_relu, np.sqrt(2))}[nonlinearity]
# 输入图片,为生成器生成1024分辨率的图像
images_in.set_shape([None, num_channels, resolution, resolution])
# 标签
labels_in.set_shape([None, label_size])
# 对输入的图片进行格式转换，一般转换为float类型
images_in = tf.cast(images_in, dtype)
labels_in = tf.cast(labels_in, dtype)
# 获取当前lod，可以简单理解为2的lod次方,代表分辨率，我们知道图片最开始输出的是低分辨率的图像，
# 虽然图像的像素都是1024，倒是经过平滑之后，几乎都看不出来是什么
lod_in = tf.cast(tf.get_variable('lod', initializer=np.float32(0.0), trainable=False), dtype)
scores_out = None
# Building blocks.
# 一个卷积之后加上一个偏置，然后经过一个激活函数，这里的res控制的是分辨率（不要和真是的分辨率混合),
# 我所说的分辨率，是吧1024按照2的res次方分割的分辨率.这里输出的格式为RGB
def fromrgb(x, res): # res = 2..resolution_log2
with tf.variable_scope('FromRGB_lod%d' % (resolution_log2 - res)):
return act(apply_bias(conv2d(x, fmaps=nf(res-1), kernel=1, gain=gain, use_wscale=use_wscale)))
# 分开了两个部分，根据res决定分辨率
def block(x, res): # res = 2..resolution_log2
with tf.variable_scope('%dx%d' % (2**res, 2**res)):
if res >= 3: # 8x8 and up
with tf.variable_scope('Conv0'):
x = act(apply_bias(conv2d(x, fmaps=nf(res-1), kernel=3, gain=gain, use_wscale=use_wscale)))
with tf.variable_scope('Conv1_down'):
x = act(apply_bias(conv2d_downscale2d(blur(x), fmaps=nf(res-2), kernel=3, gain=gain, use_wscale=use_wscale, fused_scale=fused_scale)))
else: # 4x4
if mbstd_group_size > 1:
x = minibatch_stddev_layer(x, mbstd_group_size, mbstd_num_features)
with tf.variable_scope('Conv'):
x = act(apply_bias(conv2d(x, fmaps=nf(res-1), kernel=3, gain=gain, use_wscale=use_wscale)))
with tf.variable_scope('Dense0'):
x = act(apply_bias(dense(x, fmaps=nf(res-2), gain=gain, use_wscale=use_wscale)))
with tf.variable_scope('Dense1'):
x = apply_bias(dense(x, fmaps=max(label_size, 1), gain=1, use_wscale=use_wscale))
return x
# Fixed structure: simple and efficient, but does not support progressive growing.
# 简单直接方式，直接进行搭建，分辨率固定不变，即没有分步成长
if structure == 'fixed':
x = fromrgb(images_in, resolution_log2)
for res in range(resolution_log2, 2, -1):
x = block(x, res)
scores_out = block(x, 2)
# Linear structure: simple but inefficient.
# 从高分辨率开始，逐步进行下采样，每个下采样都会有对应的RGB图像进行输出
if structure == 'linear':
img = images_in
x = fromrgb(img, resolution_log2)
for res in range(resolution_log2, 2, -1):
lod = resolution_log2 - res
x = block(x, res)
img = downscale2d(img)
y = fromrgb(img, res - 1)
with tf.variable_scope('Grow_lod%d' % lod):
x = tflib.lerp_clip(x, y, lod_in - lod)
scores_out = block(x, 2)
# Recursive structure: complex but efficient.
# 这个没有去了解，有知道的哥们可以告诉我，我会把你的解释写在这个地方
if structure == 'recursive':
def cset(cur_lambda, new_cond, new_lambda):
return lambda: tf.cond(new_cond, new_lambda, cur_lambda)
def grow(res, lod):
x = lambda: fromrgb(downscale2d(images_in, 2**lod), res)
if lod > 0: x = cset(x, (lod_in 2: y = cset(y, (lod_in > lod), lambda: tflib.lerp(x, fromrgb(downscale2d(images_in, 2**(lod+1)), res - 1), lod_in - lod))
return y()
scores_out = grow(2, resolution_log2 - 2)
# Label conditioning from "Which Training Methods for GANs do actually Converge?"
# 该处似乎为零,所以没有执行
if label_size:
with tf.variable_scope('LabelSwitch'):
scores_out = tf.reduce_sum(scores_out * labels_in, axis=1, keepdims=True)
# 这里的scores_out维度为(?,1),输出代表的应该是这个图片为真或者为假的概率
assert scores_out.dtype == tf.as_dtype(dtype)
scores_out = tf.identity(scores_out, name='scores_out')
return scores_out
这样的结构真的简单，一句总结，就是输入图片，然后通过一系列的卷积激活,全连接操作，然后得到一个值，这个值就是对应图片图片是否为真是图片的概率值。
到这里，我们对整个网络的分析就已经完成了，下面就会进入我们核心的核心了，那就是损失函数的讲解。如果觉得我写的博客对大家有所帮助，希望大家能给我点点赞，感谢大家一直以来的关注。
