Skip to content

M
models

PaddlePaddle / models
大约 1 年 前同步成功

通知 222
Star 6828
Fork 2962
M
models
未验证 提交 b0d375a8 编写于 作者: C ceci3 提交者： GitHub
浏览文件
操作

Update GAN to 1.8 (#4599) 

* fix init

* update

* update_1.8

* update paddle.reader
上级 4d7ec517
隐藏空白更改
内联 并排
Showing with 210 addition and 173 deletion
PaddleCV/gan/c_gan/c_gan.py
浏览文件 @ b0d375a8
...@@ -79,10 +79,11 @@ def train(args): ...@@ -79,10 +79,11 @@ def train(args):
g_program_test = dg_program.clone(for_test=True) g_program_test = dg_program.clone(for_test=True)
dg_logit = D_cond(g_img, conditions) dg_logit = D_cond(g_img, conditions)
dg_logit_shape = fluid.layers.shape(dg_logit)
dg_loss = loss( dg_loss = loss(
dg_logit, dg_logit,
fluid.layers.fill_constant_batch_size_like( fluid.layers.fill_constant(
input=noise, dtype='float32', shape=[-1, 1], value=1.0)) dtype='float32', shape=[dg_logit_shape[0], 1], value=1.0))
opt = fluid.optimizer.Adam(learning_rate=LEARNING_RATE) opt = fluid.optimizer.Adam(learning_rate=LEARNING_RATE)
...@@ -96,11 +97,11 @@ def train(args): ...@@ -96,11 +97,11 @@ def train(args):
exe = fluid.Executor(fluid.CUDAPlace(0)) exe = fluid.Executor(fluid.CUDAPlace(0))
exe.run(fluid.default_startup_program()) exe.run(fluid.default_startup_program())
if args.run_ce: if args.run_ce:
train_reader = paddle.batch( train_reader = fluid.io.batch(
paddle.dataset.mnist.train(), batch_size=args.batch_size) paddle.dataset.mnist.train(), batch_size=args.batch_size)
else: else:
train_reader = paddle.batch( train_reader = fluid.io.batch(
paddle.reader.shuffle( fluid.io.shuffle(
paddle.dataset.mnist.train(), buf_size=60000), paddle.dataset.mnist.train(), buf_size=60000),
batch_size=args.batch_size) batch_size=args.batch_size)
......
PaddleCV/gan/c_gan/dc_gan.py
浏览文件 @ b0d375a8
...@@ -74,10 +74,11 @@ def train(args): ...@@ -74,10 +74,11 @@ def train(args):
g_program_test = dg_program.clone(for_test=True) g_program_test = dg_program.clone(for_test=True)
dg_logit = D(g_img) dg_logit = D(g_img)
dg_logit_shape = fluid.layers.shape(dg_logit)
dg_loss = loss( dg_loss = loss(
dg_logit, dg_logit,
fluid.layers.fill_constant_batch_size_like( fluid.layers.fill_constant(
input=noise, dtype='float32', shape=[-1, 1], value=1.0)) shape=[dg_logit_shape[0], 1], dtype='float32', value=1.0))
opt = fluid.optimizer.Adam(learning_rate=LEARNING_RATE) opt = fluid.optimizer.Adam(learning_rate=LEARNING_RATE)
...@@ -92,11 +93,11 @@ def train(args): ...@@ -92,11 +93,11 @@ def train(args):
exe.run(fluid.default_startup_program()) exe.run(fluid.default_startup_program())
if args.run_ce: if args.run_ce:
train_reader = paddle.batch( train_reader = fluid.io.batch(
paddle.dataset.mnist.train(), batch_size=args.batch_size) paddle.dataset.mnist.train(), batch_size=args.batch_size)
else: else:
train_reader = paddle.batch( train_reader = fluid.io.batch(
paddle.reader.shuffle( fluid.io.shuffle(
paddle.dataset.mnist.train(), buf_size=60000), paddle.dataset.mnist.train(), buf_size=60000),
batch_size=args.batch_size) batch_size=args.batch_size)
......
PaddleCV/gan/c_gan/network.py
浏览文件 @ b0d375a8
...@@ -34,6 +34,7 @@ use_cudnn = True ...@@ -34,6 +34,7 @@ use_cudnn = True
if 'ce_mode' in os.environ: if 'ce_mode' in os.environ:
use_cudnn = False use_cudnn = False
def bn(x, name=None, act='relu'): def bn(x, name=None, act='relu'):
if name is None: if name is None:
name = get_parent_function_name() name = get_parent_function_name()
...@@ -100,8 +101,11 @@ def deconv(x, ...@@ -100,8 +101,11 @@ def deconv(x,
def conv_cond_concat(x, y): def conv_cond_concat(x, y):
"""Concatenate conditioning vector on feature map axis.""" """Concatenate conditioning vector on feature map axis."""
ones = fluid.layers.fill_constant_batch_size_like( x_shape = fluid.layers.shape(x)
x, [-1, y.shape[1], x.shape[2], x.shape[3]], "float32", 1.0) ones = fluid.layers.fill_constant(
shape=[x_shape[0], y.shape[1], x.shape[2], x.shape[3]],
dtype='float32',
value=1.0)
return fluid.layers.concat([x, ones * y], 1) return fluid.layers.concat([x, ones * y], 1)
......
PaddleCV/gan/data_reader.py
浏览文件 @ b0d375a8
...@@ -22,6 +22,7 @@ import argparse ...@@ -22,6 +22,7 @@ import argparse
import struct import struct
import os import os
import paddle import paddle
import paddle.fluid as fluid
import random import random
import sys import sys
...@@ -520,8 +521,8 @@ class data_reader(object): ...@@ -520,8 +521,8 @@ class data_reader(object):
train_labels = os.path.join(self.cfg.data_dir, self.cfg.dataset, train_labels = os.path.join(self.cfg.data_dir, self.cfg.dataset,
"train-labels-idx1-ubyte.gz") "train-labels-idx1-ubyte.gz")
train_reader = paddle.batch( train_reader = fluid.io.batch(
paddle.reader.shuffle( fluid.io.shuffle(
mnist_reader_creator(train_images, train_labels, 100), mnist_reader_creator(train_images, train_labels, 100),
buf_size=60000), buf_size=60000),
batch_size=self.cfg.batch_size) batch_size=self.cfg.batch_size)
......
PaddleCV/gan/network/AttGAN_network.py
浏览文件 @ b0d375a8
...@@ -60,9 +60,14 @@ class AttGAN_model(object): ...@@ -60,9 +60,14 @@ class AttGAN_model(object):
def concat(self, z, a): def concat(self, z, a):
"""Concatenate attribute vector on feature map axis.""" """Concatenate attribute vector on feature map axis."""
ones = fluid.layers.fill_constant_batch_size_like( batch = fluid.layers.shape(z)[0]
z, [-1, a.shape[1], z.shape[2], z.shape[3]], "float32", 1.0) ones = fluid.layers.fill_constant(
return fluid.layers.concat([z, fluid.layers.elementwise_mul(ones, a, axis=0)], axis=1) shape=[batch, a.shape[1], z.shape[2], z.shape[3]],
dtype="float32",
value=1.0)
return fluid.layers.concat(
[z, fluid.layers.elementwise_mul(
ones, a, axis=0)], axis=1)
def Genc(self, input, dim=64, n_layers=5, name='G_enc_', is_test=False): def Genc(self, input, dim=64, n_layers=5, name='G_enc_', is_test=False):
z = input z = input
......
PaddleCV/gan/network/STGAN_network.py
浏览文件 @ b0d375a8
...@@ -82,9 +82,14 @@ class STGAN_model(object): ...@@ -82,9 +82,14 @@ class STGAN_model(object):
def concat(self, z, a): def concat(self, z, a):
"""Concatenate attribute vector on feature map axis.""" """Concatenate attribute vector on feature map axis."""
ones = fluid.layers.fill_constant_batch_size_like( batch = fluid.layers.shape(z)[0]
z, [-1, a.shape[1], z.shape[2], z.shape[3]], "float32", 1.0) ones = fluid.layers.fill_constant(
return fluid.layers.concat([z, fluid.layers.elementwise_mul(ones, a, axis=0)], axis=1) shape=[batch, a.shape[1], z.shape[2], z.shape[3]],
dtype="float32",
value=1.0)
return fluid.layers.concat(
[z, fluid.layers.elementwise_mul(
ones, a, axis=0)], axis=1)
def Genc(self, input, dim=64, n_layers=5, name='G_enc_', is_test=False): def Genc(self, input, dim=64, n_layers=5, name='G_enc_', is_test=False):
z = input z = input
......
PaddleCV/gan/network/base_network.py
浏览文件 @ b0d375a8
...@@ -42,7 +42,9 @@ def norm_layer(input, ...@@ -42,7 +42,9 @@ def norm_layer(input,
if norm_type == 'batch_norm': if norm_type == 'batch_norm':
if affine == True: if affine == True:
param_attr = fluid.ParamAttr( param_attr = fluid.ParamAttr(
name=name + '_w', initializer=fluid.initializer.Normal(loc=1.0, scale=0.02)) name=name + '_w',
initializer=fluid.initializer.Normal(
loc=1.0, scale=0.02))
bias_attr = fluid.ParamAttr( bias_attr = fluid.ParamAttr(
name=name + '_b', name=name + '_b',
initializer=fluid.initializer.Constant(value=0.0)) initializer=fluid.initializer.Constant(value=0.0))
...@@ -366,8 +368,11 @@ def linear(input, ...@@ -366,8 +368,11 @@ def linear(input,
def conv_cond_concat(x, y): def conv_cond_concat(x, y):
ones = fluid.layers.fill_constant_batch_size_like( batch = fluid.layers.shape(x)[0]
x, [-1, y.shape[1], x.shape[2], x.shape[3]], "float32", 1.0) ones = fluid.layers.fill_constant(
shape=[ones, y.shape[1], x.shape[2], x.shape[3]],
dtype="float32",
value=1.0)
out = fluid.layers.concat([x, ones * y], 1) out = fluid.layers.concat([x, ones * y], 1)
return out return out
......
PaddleCV/gan/trainer/AttGAN.py
浏览文件 @ b0d375a8
...@@ -44,11 +44,9 @@ class GTrainer(): ...@@ -44,11 +44,9 @@ class GTrainer():
self.g_loss_fake = -1 * fluid.layers.mean(self.pred_fake) self.g_loss_fake = -1 * fluid.layers.mean(self.pred_fake)
#lsgan #lsgan
elif cfg.gan_mode == "lsgan": elif cfg.gan_mode == "lsgan":
ones = fluid.layers.fill_constant_batch_size_like( fake_shape = fluid.layers.shape(self.pred_fake)
input=self.pred_fake, ones = fluid.layers.fill_constant(
shape=self.pred_fake.shape, shape=fake_shape, value=1.0, dtype='float32')
value=1.0,
dtype='float32')
self.g_loss_fake = fluid.layers.mean( self.g_loss_fake = fluid.layers.mean(
fluid.layers.square( fluid.layers.square(
fluid.layers.elementwise_sub( fluid.layers.elementwise_sub(
...@@ -106,11 +104,9 @@ class DTrainer(): ...@@ -106,11 +104,9 @@ class DTrainer():
self.d_loss = self.d_loss_real + self.d_loss_fake + 1.0 * self.d_loss_cls + cfg.lambda_gp * self.d_loss_gp self.d_loss = self.d_loss_real + self.d_loss_fake + 1.0 * self.d_loss_cls + cfg.lambda_gp * self.d_loss_gp
#lsgan #lsgan
elif cfg.gan_mode == "lsgan": elif cfg.gan_mode == "lsgan":
ones = fluid.layers.fill_constant_batch_size_like( real_shape = fluid.layers.shape(self.pred_real)
input=self.pred_real, ones = fluid.layers.fill_constant(
shape=self.pred_real.shape, shape=real_shape, value=1.0, dtype='float32')
value=1.0,
dtype='float32')
self.d_loss_real = fluid.layers.mean( self.d_loss_real = fluid.layers.mean(
fluid.layers.square( fluid.layers.square(
fluid.layers.elementwise_sub( fluid.layers.elementwise_sub(
...@@ -145,31 +141,31 @@ class DTrainer(): ...@@ -145,31 +141,31 @@ class DTrainer():
def gradient_penalty(self, f, real, fake=None, cfg=None, name=None): def gradient_penalty(self, f, real, fake=None, cfg=None, name=None):
def _interpolate(a, b=None): def _interpolate(a, b=None):
a_shape = fluid.layers.shape(a)
if b is None: if b is None:
if cfg.enable_ce: if cfg.enable_ce:
beta = fluid.layers.uniform_random_batch_size_like( beta = fluid.layers.uniform_random(
input=a, shape=a.shape, min=0.0, max=1.0, seed=1) shape=a_shape, min=0.0, max=1.0, seed=1)
else: else:
beta = fluid.layers.uniform_random_batch_size_like( beta = fluid.layers.uniform_random(
input=a, shape=a.shape, min=0.0, max=1.0) shape=a_shape, min=0.0, max=1.0)
mean = fluid.layers.reduce_mean( mean = fluid.layers.reduce_mean(
a, dim=list(range(len(a.shape))), keep_dim=True) a, dim=list(range(len(a.shape))))
input_sub_mean = fluid.layers.elementwise_sub(a, mean, axis=0) input_sub_mean = fluid.layers.elementwise_sub(a, mean, axis=0)
var = fluid.layers.reduce_mean( var = fluid.layers.reduce_mean(
fluid.layers.square(input_sub_mean), fluid.layers.square(input_sub_mean),
dim=list(range(len(a.shape))), dim=list(range(len(a.shape))))
keep_dim=True)
b = beta * fluid.layers.sqrt(var) * 0.5 + a b = beta * fluid.layers.sqrt(var) * 0.5 + a
shape = [a.shape[0]] shape = [a.shape[0]]
if cfg.enable_ce: if cfg.enable_ce:
alpha = fluid.layers.uniform_random_batch_size_like( alpha = fluid.layers.uniform_random(
input=a, shape=shape, min=0.0, max=1.0, seed=1) shape=a_shape[0], min=0.0, max=1.0, seed=1)
else: else:
alpha = fluid.layers.uniform_random_batch_size_like( alpha = fluid.layers.uniform_random(
input=a, shape=shape, min=0.0, max=1.0) shape=a_shape[0], min=0.0, max=1.0)
inner = fluid.layers.elementwise_mul((b-a), alpha, axis=0) + a inner = fluid.layers.elementwise_mul((b - a), alpha, axis=0) + a
return inner return inner
x = _interpolate(real, fake) x = _interpolate(real, fake)
...@@ -336,7 +332,7 @@ class AttGAN(object): ...@@ -336,7 +332,7 @@ class AttGAN(object):
if self.cfg.enable_ce: if self.cfg.enable_ce:
gen_trainer_program.random_seed = 90 gen_trainer_program.random_seed = 90
dis_trainer_program.random_seed = 90 dis_trainer_program.random_seed = 90
t_time = 0 t_time = 0
for epoch_id in range(self.cfg.epoch): for epoch_id in range(self.cfg.epoch):
...@@ -379,7 +375,7 @@ class AttGAN(object): ...@@ -379,7 +375,7 @@ class AttGAN(object):
sys.stdout.flush() sys.stdout.flush()
batch_id += 1 batch_id += 1
if self.cfg.enable_ce and batch_id == 100: if self.cfg.enable_ce and batch_id == 100:
break break
if self.cfg.run_test: if self.cfg.run_test:
image_name = fluid.data( image_name = fluid.data(
...@@ -402,17 +398,23 @@ class AttGAN(object): ...@@ -402,17 +398,23 @@ class AttGAN(object):
test_loader) test_loader)
if self.cfg.save_checkpoints: if self.cfg.save_checkpoints:
utility.checkpoints(epoch_id, self.cfg, gen_trainer, utility.checkpoints(epoch_id, self.cfg, gen_trainer, "net_G")
"net_G") utility.checkpoints(epoch_id, self.cfg, dis_trainer, "net_D")
utility.checkpoints(epoch_id, self.cfg, dis_trainer,
"net_D")
# used for continuous evaluation # used for continuous evaluation
if self.cfg.enable_ce: if self.cfg.enable_ce:
device_num = fluid.core.get_cuda_device_count() if self.cfg.use_gpu else 1 device_num = fluid.core.get_cuda_device_count(
print("kpis\tattgan_g_loss_fake_card{}\t{}".format(device_num, g_loss_fake[0])) ) if self.cfg.use_gpu else 1
print("kpis\tattgan_g_loss_rec_card{}\t{}".format(device_num, g_loss_rec[0])) print("kpis\tattgan_g_loss_fake_card{}\t{}".format(
print("kpis\tattgan_g_loss_cls_card{}\t{}".format(device_num, g_loss_cls[0])) device_num, g_loss_fake[0]))
print("kpis\tattgan_d_loss_real_card{}\t{}".format(device_num, d_loss_real[0])) print("kpis\tattgan_g_loss_rec_card{}\t{}".format(
print("kpis\tattgan_d_loss_fake_card{}\t{}".format(device_num,d_loss_fake[0])) device_num, g_loss_rec[0]))
print("kpis\tattgan_d_loss_gp_card{}\t{}".format(device_num,d_loss_gp[0])) print("kpis\tattgan_g_loss_cls_card{}\t{}".format(
print("kpis\tattgan_Batch_time_cost_card{}\t{}".format(device_num,batch_time)) device_num, g_loss_cls[0]))
print("kpis\tattgan_d_loss_real_card{}\t{}".format(
device_num, d_loss_real[0]))
print("kpis\tattgan_d_loss_fake_card{}\t{}".format(
device_num, d_loss_fake[0]))
print("kpis\tattgan_d_loss_gp_card{}\t{}".format(device_num,
d_loss_gp[0]))
print("kpis\tattgan_Batch_time_cost_card{}\t{}".format(
device_num, batch_time))
PaddleCV/gan/trainer/CGAN.py
浏览文件 @ b0d375a8
...@@ -37,8 +37,9 @@ class GTrainer(): ...@@ -37,8 +37,9 @@ class GTrainer():
self.fake = model.network_G(input, conditions, name="G") self.fake = model.network_G(input, conditions, name="G")
self.infer_program = self.program.clone(for_test=True) self.infer_program = self.program.clone(for_test=True)
d_fake = model.network_D(self.fake, conditions, name="D") d_fake = model.network_D(self.fake, conditions, name="D")
fake_labels = fluid.layers.fill_constant_batch_size_like( batch = fluid.layers.shape(input)[0]
input=input, dtype='float32', shape=[-1, 1], value=1.0) fake_labels = fluid.layers.fill_constant(
dtype='float32', shape=[batch, 1], value=1.0)
self.g_loss = fluid.layers.reduce_mean( self.g_loss = fluid.layers.reduce_mean(
fluid.layers.sigmoid_cross_entropy_with_logits( fluid.layers.sigmoid_cross_entropy_with_logits(
x=d_fake, label=fake_labels)) x=d_fake, label=fake_labels))
......
PaddleCV/gan/trainer/DCGAN.py
浏览文件 @ b0d375a8
...@@ -38,8 +38,9 @@ class GTrainer(): ...@@ -38,8 +38,9 @@ class GTrainer():
self.fake = model.network_G(input, name='G') self.fake = model.network_G(input, name='G')
self.infer_program = self.program.clone(for_test=True) self.infer_program = self.program.clone(for_test=True)
d_fake = model.network_D(self.fake, name="D") d_fake = model.network_D(self.fake, name="D")
fake_labels = fluid.layers.fill_constant_batch_size_like( batch = fluid.layers.shape(input)[0]
input, dtype='float32', shape=[-1, 1], value=1.0) fake_labels = fluid.layers.fill_constant(
dtype='float32', shape=[batch, 1], value=1.0)
self.g_loss = fluid.layers.reduce_mean( self.g_loss = fluid.layers.reduce_mean(
fluid.layers.sigmoid_cross_entropy_with_logits( fluid.layers.sigmoid_cross_entropy_with_logits(
x=d_fake, label=fake_labels)) x=d_fake, label=fake_labels))
......
PaddleCV/gan/trainer/Pix2pix.py
浏览文件 @ b0d375a8
...@@ -33,10 +33,10 @@ class GTrainer(): ...@@ -33,10 +33,10 @@ class GTrainer():
self.infer_program = self.program.clone() self.infer_program = self.program.clone()
AB = fluid.layers.concat([input_A, self.fake_B], 1) AB = fluid.layers.concat([input_A, self.fake_B], 1)
self.pred = model.network_D(AB, "discriminator", cfg) self.pred = model.network_D(AB, "discriminator", cfg)
batch = fluid.layers.shape(self.pred)[0]
if cfg.gan_mode == "lsgan": if cfg.gan_mode == "lsgan":
ones = fluid.layers.fill_constant_batch_size_like( ones = fluid.layers.fill_constant(
input=self.pred, shape=[batch] + list(self.pred.shape[1:]),
shape=self.pred.shape,
value=1, value=1,
dtype='float32') dtype='float32')
self.g_loss_gan = fluid.layers.reduce_mean( self.g_loss_gan = fluid.layers.reduce_mean(
...@@ -49,9 +49,8 @@ class GTrainer(): ...@@ -49,9 +49,8 @@ class GTrainer():
self.pred, self.pred,
[-1, pred_shape[1] * pred_shape[2] * pred_shape[3]], [-1, pred_shape[1] * pred_shape[2] * pred_shape[3]],
inplace=True) inplace=True)
ones = fluid.layers.fill_constant_batch_size_like( ones = fluid.layers.fill_constant(
input=self.pred, shape=[batch] + list(self.pred.shape[1:]),
shape=self.pred.shape,
value=1, value=1,
dtype='float32') dtype='float32')
self.g_loss_gan = fluid.layers.mean( self.g_loss_gan = fluid.layers.mean(
...@@ -106,10 +105,10 @@ class DTrainer(): ...@@ -106,10 +105,10 @@ class DTrainer():
self.real_AB, "discriminator", cfg=cfg) self.real_AB, "discriminator", cfg=cfg)
self.pred_fake = model.network_D( self.pred_fake = model.network_D(
self.fake_AB, "discriminator", cfg=cfg) self.fake_AB, "discriminator", cfg=cfg)
batch = fluid.layers.shape(input_A)[0]
if cfg.gan_mode == "lsgan": if cfg.gan_mode == "lsgan":
ones = fluid.layers.fill_constant_batch_size_like( ones = fluid.layers.fill_constant(
input=self.pred_real, shape=[batch] + list(self.pred_real.shape[1:]),
shape=self.pred_real.shape,
value=1, value=1,
dtype='float32') dtype='float32')
self.d_loss_real = fluid.layers.reduce_mean( self.d_loss_real = fluid.layers.reduce_mean(
...@@ -128,14 +127,12 @@ class DTrainer(): ...@@ -128,14 +127,12 @@ class DTrainer():
self.pred_fake, self.pred_fake,
[-1, pred_shape[1] * pred_shape[2] * pred_shape[3]], [-1, pred_shape[1] * pred_shape[2] * pred_shape[3]],
inplace=True) inplace=True)
zeros = fluid.layers.fill_constant_batch_size_like( zeros = fluid.layers.fill_constant(
input=self.pred_fake, shape=[batch] + list(self.pred_fake.shape[1:]),
shape=self.pred_fake.shape,
value=0, value=0,
dtype='float32') dtype='float32')
ones = fluid.layers.fill_constant_batch_size_like( ones = fluid.layers.fill_constant(
input=self.pred_real, shape=[batch] + list(self.pred_real.shape[1:]),
shape=self.pred_real.shape,
value=1, value=1,
dtype='float32') dtype='float32')
self.d_loss_real = fluid.layers.mean( self.d_loss_real = fluid.layers.mean(
...@@ -283,7 +280,8 @@ class Pix2pix(object): ...@@ -283,7 +280,8 @@ class Pix2pix(object):
devices_num = utility.get_device_num(self.cfg) devices_num = utility.get_device_num(self.cfg)
fake_per_device = int(len(fake_B_tmp) / devices_num) fake_per_device = int(len(fake_B_tmp) / devices_num)
for dev in range(devices_num): for dev in range(devices_num):
tensor[dev]['input_fake'] = fake_B_tmp[dev * fake_per_device : (dev+1) * fake_per_device] tensor[dev]['input_fake'] = fake_B_tmp[
dev * fake_per_device:(dev + 1) * fake_per_device]
# optimize the discriminator network # optimize the discriminator network
d_loss_real, d_loss_fake = exe.run(dis_trainer_program, d_loss_real, d_loss_fake = exe.run(dis_trainer_program,
...@@ -338,10 +336,8 @@ class Pix2pix(object): ...@@ -338,10 +336,8 @@ class Pix2pix(object):
A_id2name=self.id2name) A_id2name=self.id2name)
if self.cfg.save_checkpoints: if self.cfg.save_checkpoints:
utility.checkpoints(epoch_id, self.cfg, gen_trainer, utility.checkpoints(epoch_id, self.cfg, gen_trainer, "net_G")
"net_G") utility.checkpoints(epoch_id, self.cfg, dis_trainer, "net_D")
utility.checkpoints(epoch_id, self.cfg, dis_trainer,
"net_D")
if self.cfg.enable_ce: if self.cfg.enable_ce:
device_num = fluid.core.get_cuda_device_count( device_num = fluid.core.get_cuda_device_count(
) if self.cfg.use_gpu else 1 ) if self.cfg.use_gpu else 1
......
PaddleCV/gan/trainer/SPADE.py
浏览文件 @ b0d375a8
...@@ -144,11 +144,9 @@ class DTrainer(): ...@@ -144,11 +144,9 @@ class DTrainer():
#####gan loss #####gan loss
self.gan_loss_fake = 0 self.gan_loss_fake = 0
for pred_i in self.pred_fake: for pred_i in self.pred_fake:
zeros = fluid.layers.fill_constant_batch_size_like( pred_shape = fluid.layers.shape(pred_i[-1])
input=pred_i[-1], zeros = fluid.layers.fill_constant(
shape=pred_i[-1].shape, shape=pred_shape, value=0, dtype='float32')
value=0,
dtype='float32')
if isinstance(pred_i, list): if isinstance(pred_i, list):
pred_i = pred_i[-1] pred_i = pred_i[-1]
minval = fluid.layers.elementwise_min(-1 * pred_i - 1, zeros) minval = fluid.layers.elementwise_min(-1 * pred_i - 1, zeros)
...@@ -158,11 +156,9 @@ class DTrainer(): ...@@ -158,11 +156,9 @@ class DTrainer():
self.gan_loss_real = 0 self.gan_loss_real = 0
for pred_i in self.pred_real: for pred_i in self.pred_real:
zeros = fluid.layers.fill_constant_batch_size_like( pred_shape = fluid.layers.shape(pred_i[-1])
input=pred_i[-1], zeros = fluid.layers.fill_constant(
shape=pred_i[-1].shape, shape=pred_shape, value=0, dtype='float32')
value=0,
dtype='float32')
if isinstance(pred_i, list): if isinstance(pred_i, list):
pred_i = pred_i[-1] pred_i = pred_i[-1]
minval = fluid.layers.elementwise_min(pred_i - 1, zeros) minval = fluid.layers.elementwise_min(pred_i - 1, zeros)
...@@ -298,7 +294,7 @@ class SPADE(object): ...@@ -298,7 +294,7 @@ class SPADE(object):
name='input_fake', shape=data_shape, dtype='float32') name='input_fake', shape=data_shape, dtype='float32')
# used for continuous evaluation # used for continuous evaluation
if self.cfg.enable_ce: if self.cfg.enable_ce:
fluid.default_startup_program().random_seed = 90 fluid.default_startup_program().random_seed = 90
gen_trainer = GTrainer(input_A, input_B, input_C, self.cfg, gen_trainer = GTrainer(input_A, input_B, input_C, self.cfg,
self.batch_num) self.batch_num)
...@@ -348,7 +344,7 @@ class SPADE(object): ...@@ -348,7 +344,7 @@ class SPADE(object):
if self.cfg.enable_ce: if self.cfg.enable_ce:
gen_trainer_program.random_seed = 90 gen_trainer_program.random_seed = 90
dis_trainer_program.random_seed = 90 dis_trainer_program.random_seed = 90
t_time = 0 t_time = 0
for epoch_id in range(self.cfg.epoch): for epoch_id in range(self.cfg.epoch):
...@@ -422,16 +418,21 @@ class SPADE(object): ...@@ -422,16 +418,21 @@ class SPADE(object):
A_id2name=self.id2name) A_id2name=self.id2name)
if self.cfg.save_checkpoints: if self.cfg.save_checkpoints:
utility.checkpoints(epoch_id, self.cfg, gen_trainer, utility.checkpoints(epoch_id, self.cfg, gen_trainer, "net_G")
"net_G") utility.checkpoints(epoch_id, self.cfg, dis_trainer, "net_D")
utility.checkpoints(epoch_id, self.cfg, dis_trainer,
"net_D")
# used for continuous evaluation # used for continuous evaluation
if self.cfg.enable_ce: if self.cfg.enable_ce:
device_num = fluid.core.get_cuda_device_count() if self.cfg.use_gpu else 1 device_num = fluid.core.get_cuda_device_count(
print("kpis\tspade_g_loss_gan_card{}\t{}".format(device_num, g_loss_gan[0])) ) if self.cfg.use_gpu else 1
print("kpis\tspade_g_loss_vgg_card{}\t{}".format(device_num,g_loss_vgg[0])) print("kpis\tspade_g_loss_gan_card{}\t{}".format(device_num,
print("kpis\tspade_g_loss_feat_card{}\t{}".format(device_num,g_loss_feat[0])) g_loss_gan[0]))
print("kpis\tspade_d_loss_real_card{}\t{}".format(device_num,d_loss_real[0])) print("kpis\tspade_g_loss_vgg_card{}\t{}".format(device_num,
print("kpis\tspade_d_loss_fake_card{}\t{}".format(device_num,d_loss_fake[0])) g_loss_vgg[0]))
print("kpis\tspade_Batch_time_cost_card{}\t{}".format(device_num,batch_time)) print("kpis\tspade_g_loss_feat_card{}\t{}".format(
device_num, g_loss_feat[0]))
print("kpis\tspade_d_loss_real_card{}\t{}".format(
device_num, d_loss_real[0]))
print("kpis\tspade_d_loss_fake_card{}\t{}".format(
device_num, d_loss_fake[0]))
print("kpis\tspade_Batch_time_cost_card{}\t{}".format(
device_num, batch_time))
PaddleCV/gan/trainer/STGAN.py
浏览文件 @ b0d375a8
...@@ -45,11 +45,9 @@ class GTrainer(): ...@@ -45,11 +45,9 @@ class GTrainer():
self.g_loss_fake = -1 * fluid.layers.mean(self.pred_fake) self.g_loss_fake = -1 * fluid.layers.mean(self.pred_fake)
#lsgan #lsgan
elif cfg.gan_mode == "lsgan": elif cfg.gan_mode == "lsgan":
ones = fluid.layers.fill_constant_batch_size_like( fake_shape = fluid.layers.shape(self.pred_fake)
input=self.pred_fake, ones = fluid.layers.fill_constant(
shape=self.pred_fake.shape, shape=fake_shape, value=1.0, dtype='float32')
value=1.0,
dtype='float32')
self.g_loss_fake = fluid.layers.mean( self.g_loss_fake = fluid.layers.mean(
fluid.layers.square( fluid.layers.square(
fluid.layers.elementwise_sub( fluid.layers.elementwise_sub(
...@@ -108,11 +106,9 @@ class DTrainer(): ...@@ -108,11 +106,9 @@ class DTrainer():
self.d_loss = self.d_loss_real + self.d_loss_fake + 1.0 * self.d_loss_cls + cfg.lambda_gp * self.d_loss_gp self.d_loss = self.d_loss_real + self.d_loss_fake + 1.0 * self.d_loss_cls + cfg.lambda_gp * self.d_loss_gp
#lsgan #lsgan
elif cfg.gan_mode == "lsgan": elif cfg.gan_mode == "lsgan":
ones = fluid.layers.fill_constant_batch_size_like( real_shape = fluid.layers.shape(self.pred_real)
input=self.pred_real, ones = fluid.layers.fill_constant(
shape=self.pred_real.shape, shape=real_shape, value=1.0, dtype='float32')
value=1.0,
dtype='float32')
self.d_loss_real = fluid.layers.mean( self.d_loss_real = fluid.layers.mean(
fluid.layers.square( fluid.layers.square(
fluid.layers.elementwise_sub( fluid.layers.elementwise_sub(
...@@ -149,31 +145,30 @@ class DTrainer(): ...@@ -149,31 +145,30 @@ class DTrainer():
def gradient_penalty(self, f, real, fake=None, cfg=None, name=None): def gradient_penalty(self, f, real, fake=None, cfg=None, name=None):
def _interpolate(a, b=None): def _interpolate(a, b=None):
a_shape = fluid.layers.shape(a)
if b is None: if b is None:
if cfg.enable_ce: if cfg.enable_ce:
beta = fluid.layers.uniform_random_batch_size_like( beta = fluid.layers.uniform_random(
input=a, shape=a.shape, min=0.0, max=1.0, seed=1) shape=a_shape, min=0.0, max=1.0, seed=1)
else: else:
beta = fluid.layers.uniform_random_batch_size_like( beta = fluid.layers.uniform_random(
input=a, shape=a.shape, min=0.0, max=1.0) shape=a_shape, min=0.0, max=1.0)
mean = fluid.layers.reduce_mean( mean = fluid.layers.reduce_mean(
a, dim=list(range(len(a.shape))), keep_dim=True) a, dim=list(range(len(a.shape))))
input_sub_mean = fluid.layers.elementwise_sub(a, mean, axis=0) input_sub_mean = fluid.layers.elementwise_sub(a, mean, axis=0)
var = fluid.layers.reduce_mean( var = fluid.layers.reduce_mean(
fluid.layers.square(input_sub_mean), fluid.layers.square(input_sub_mean),
dim=list(range(len(a.shape))), dim=list(range(len(a.shape))))
keep_dim=True)
b = beta * fluid.layers.sqrt(var) * 0.5 + a b = beta * fluid.layers.sqrt(var) * 0.5 + a
shape = [a.shape[0]]
if cfg.enable_ce: if cfg.enable_ce:
alpha = fluid.layers.uniform_random_batch_size_like( alpha = fluid.layers.uniform_random(
input=a, shape=shape, min=0.0, max=1.0, seed=1) shape=a_shape[0], min=0.0, max=1.0, seed=1)
else: else:
alpha = fluid.layers.uniform_random_batch_size_like( alpha = fluid.layers.uniform_random(
input=a, shape=shape, min=0.0, max=1.0) shape=a_shape[0], min=0.0, max=1.0)
inner = fluid.layers.elementwise_mul((b-a), alpha, axis=0) + a inner = fluid.layers.elementwise_mul((b - a), alpha, axis=0) + a
return inner return inner
x = _interpolate(real, fake) x = _interpolate(real, fake)
...@@ -221,7 +216,10 @@ class STGAN(object): ...@@ -221,7 +216,10 @@ class STGAN(object):
default=1024, default=1024,
help="the base fc dim in discriminator") help="the base fc dim in discriminator")
parser.add_argument( parser.add_argument(
'--use_gru', type=ast.literal_eval, default=True, help="whether to use GRU") '--use_gru',
type=ast.literal_eval,
default=True,
help="whether to use GRU")
parser.add_argument( parser.add_argument(
'--lambda_cls', '--lambda_cls',
type=float, type=float,
...@@ -345,7 +343,7 @@ class STGAN(object): ...@@ -345,7 +343,7 @@ class STGAN(object):
if self.cfg.enable_ce: if self.cfg.enable_ce:
gen_trainer_program.random_seed = 90 gen_trainer_program.random_seed = 90
dis_trainer_program.random_seed = 90 dis_trainer_program.random_seed = 90
t_time = 0 t_time = 0
total_train_batch = 0 # used for benchmark total_train_batch = 0 # used for benchmark
...@@ -353,7 +351,7 @@ class STGAN(object): ...@@ -353,7 +351,7 @@ class STGAN(object):
for epoch_id in range(self.cfg.epoch): for epoch_id in range(self.cfg.epoch):
batch_id = 0 batch_id = 0
for data in loader(): for data in loader():
if self.cfg.max_iter and total_train_batch == self.cfg.max_iter: # used for benchmark if self.cfg.max_iter and total_train_batch == self.cfg.max_iter: # used for benchmark
return return
s_time = time.time() s_time = time.time()
# optimize the discriminator network # optimize the discriminator network
...@@ -389,7 +387,7 @@ class STGAN(object): ...@@ -389,7 +387,7 @@ class STGAN(object):
sys.stdout.flush() sys.stdout.flush()
batch_id += 1 batch_id += 1
if self.cfg.enable_ce and batch_id == 100: if self.cfg.enable_ce and batch_id == 100:
break break
total_train_batch += 1 # used for benchmark total_train_batch += 1 # used for benchmark
# profiler tools # profiler tools
...@@ -418,19 +416,27 @@ class STGAN(object): ...@@ -418,19 +416,27 @@ class STGAN(object):
test_loader) test_loader)
if self.cfg.save_checkpoints: if self.cfg.save_checkpoints:
utility.checkpoints(epoch_id, self.cfg, gen_trainer, utility.checkpoints(epoch_id, self.cfg, gen_trainer, "net_G")
"net_G") utility.checkpoints(epoch_id, self.cfg, dis_trainer, "net_D")
utility.checkpoints(epoch_id, self.cfg, dis_trainer,
"net_D")
# used for continuous evaluation # used for continuous evaluation
if self.cfg.enable_ce: if self.cfg.enable_ce:
device_num = fluid.core.get_cuda_device_count() if self.cfg.use_gpu else 1 device_num = fluid.core.get_cuda_device_count(
print("kpis\tstgan_g_loss_fake_card{}\t{}".format(device_num, g_loss_fake[0])) ) if self.cfg.use_gpu else 1
print("kpis\tstgan_g_loss_rec_card{}\t{}".format(device_num, g_loss_rec[0])) print("kpis\tstgan_g_loss_fake_card{}\t{}".format(
print("kpis\tstgan_g_loss_cls_card{}\t{}".format(device_num, g_loss_cls[0])) device_num, g_loss_fake[0]))
print("kpis\tstgan_d_loss_card{}\t{}".format(device_num, d_loss[0])) print("kpis\tstgan_g_loss_rec_card{}\t{}".format(device_num,
print("kpis\tstgan_d_loss_real_card{}\t{}".format(device_num, d_loss_real[0])) g_loss_rec[0]))
print("kpis\tstgan_d_loss_fake_card{}\t{}".format(device_num,d_loss_fake[0])) print("kpis\tstgan_g_loss_cls_card{}\t{}".format(device_num,
print("kpis\tstgan_d_loss_cls_card{}\t{}".format(device_num, d_loss_cls[0])) g_loss_cls[0]))
print("kpis\tstgan_d_loss_gp_card{}\t{}".format(device_num,d_loss_gp[0])) print("kpis\tstgan_d_loss_card{}\t{}".format(device_num, d_loss[
print("kpis\tstgan_Batch_time_cost_card{}\t{}".format(device_num,batch_time)) 0]))
print("kpis\tstgan_d_loss_real_card{}\t{}".format(
device_num, d_loss_real[0]))
print("kpis\tstgan_d_loss_fake_card{}\t{}".format(
device_num, d_loss_fake[0]))
print("kpis\tstgan_d_loss_cls_card{}\t{}".format(device_num,
d_loss_cls[0]))
print("kpis\tstgan_d_loss_gp_card{}\t{}".format(device_num,
d_loss_gp[0]))
print("kpis\tstgan_Batch_time_cost_card{}\t{}".format(
device_num, batch_time))
PaddleCV/gan/trainer/StarGAN.py
浏览文件 @ b0d375a8
...@@ -149,15 +149,17 @@ class DTrainer(): ...@@ -149,15 +149,17 @@ class DTrainer():
def gradient_penalty(self, f, real, fake, cfg=None, name=None): def gradient_penalty(self, f, real, fake, cfg=None, name=None):
def _interpolate(a, b): def _interpolate(a, b):
shape = [a.shape[0]] a_shape = fluid.layers.shape(a)
if cfg.enable_ce: if cfg.enable_ce:
alpha = fluid.layers.uniform_random_batch_size_like( alpha = fluid.layers.uniform_random(
input=a, shape=shape, min=0.0, max=1.0, seed=1) shape=[a_shape[0]], min=0.0, max=1.0, seed=1)
else: else:
alpha = fluid.layers.uniform_random_batch_size_like( alpha = fluid.layers.uniform_random(
input=a, shape=shape, min=0.0, max=1.0) shape=[a_shape[0]], min=0.0, max=1.0)
inner = fluid.layers.elementwise_mul(b, (1.0-alpha), axis=0) + fluid.layers.elementwise_mul(a, alpha, axis=0) inner = fluid.layers.elementwise_mul(
b, (1.0 - alpha), axis=0) + fluid.layers.elementwise_mul(
a, alpha, axis=0)
return inner return inner
x = _interpolate(real, fake) x = _interpolate(real, fake)
...@@ -316,7 +318,7 @@ class StarGAN(object): ...@@ -316,7 +318,7 @@ class StarGAN(object):
for epoch_id in range(self.cfg.epoch): for epoch_id in range(self.cfg.epoch):
batch_id = 0 batch_id = 0
for data in loader(): for data in loader():
if self.cfg.max_iter and total_train_batch == self.cfg.max_iter: # used for benchmark if self.cfg.max_iter and total_train_batch == self.cfg.max_iter: # used for benchmark
return return
s_time = time.time() s_time = time.time()
d_loss_real, d_loss_fake, d_loss, d_loss_cls, d_loss_gp = exe.run( d_loss_real, d_loss_fake, d_loss, d_loss_cls, d_loss_gp = exe.run(
...@@ -355,7 +357,7 @@ class StarGAN(object): ...@@ -355,7 +357,7 @@ class StarGAN(object):
batch_id += 1 batch_id += 1
# used for ce # used for ce
if self.cfg.enable_ce and batch_id == 100: if self.cfg.enable_ce and batch_id == 100:
break break
total_train_batch += 1 # used for benchmark total_train_batch += 1 # used for benchmark
# profiler tools # profiler tools
...@@ -380,22 +382,28 @@ class StarGAN(object): ...@@ -380,22 +382,28 @@ class StarGAN(object):
if self.cfg.use_gpu else fluid.cpu_places()) if self.cfg.use_gpu else fluid.cpu_places())
test_program = gen_trainer.infer_program test_program = gen_trainer.infer_program
utility.save_test_image(epoch_id, self.cfg, exe, place, utility.save_test_image(epoch_id, self.cfg, exe, place,
test_program, gen_trainer, test_program, gen_trainer, test_loader)
test_loader)
if self.cfg.save_checkpoints: if self.cfg.save_checkpoints:
utility.checkpoints(epoch_id, self.cfg, gen_trainer, utility.checkpoints(epoch_id, self.cfg, gen_trainer, "net_G")
"net_G") utility.checkpoints(epoch_id, self.cfg, dis_trainer, "net_D")
utility.checkpoints(epoch_id, self.cfg, dis_trainer,
"net_D")
# used for continuous evaluation # used for continuous evaluation
if self.cfg.enable_ce: if self.cfg.enable_ce:
device_num = fluid.core.get_cuda_device_count() if self.cfg.use_gpu else 1 device_num = fluid.core.get_cuda_device_count(
print("kpis\tstargan_g_loss_fake_card{}\t{}".format(device_num, g_loss_fake[0])) ) if self.cfg.use_gpu else 1
print("kpis\tstargan_g_loss_rec_card{}\t{}".format(device_num, g_loss_rec[0])) print("kpis\tstargan_g_loss_fake_card{}\t{}".format(
print("kpis\tstargan_g_loss_cls_card{}\t{}".format(device_num, g_loss_cls[0])) device_num, g_loss_fake[0]))
print("kpis\tstargan_d_loss_real_card{}\t{}".format(device_num, d_loss_real[0])) print("kpis\tstargan_g_loss_rec_card{}\t{}".format(
print("kpis\tstargan_d_loss_fake_card{}\t{}".format(device_num,d_loss_fake[0])) device_num, g_loss_rec[0]))
print("kpis\tstargan_d_loss_cls_card{}\t{}".format(device_num, d_loss_cls[0])) print("kpis\tstargan_g_loss_cls_card{}\t{}".format(
print("kpis\tstargan_d_loss_gp_card{}\t{}".format(device_num,d_loss_gp[0])) device_num, g_loss_cls[0]))
print("kpis\tstargan_Batch_time_cost_card{}\t{}".format(device_num,batch_time)) print("kpis\tstargan_d_loss_real_card{}\t{}".format(
device_num, d_loss_real[0]))
print("kpis\tstargan_d_loss_fake_card{}\t{}".format(
device_num, d_loss_fake[0]))
print("kpis\tstargan_d_loss_cls_card{}\t{}".format(
device_num, d_loss_cls[0]))
print("kpis\tstargan_d_loss_gp_card{}\t{}".format(device_num,
d_loss_gp[0]))
print("kpis\tstargan_Batch_time_cost_card{}\t{}".format(
device_num, batch_time))
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或