Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Fix imagenet models for v2 #104

Merged
merged 1 commit into from
Aug 17, 2017
Merged

Fix imagenet models for v2 #104

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
20 changes: 10 additions & 10 deletions examples/imagenet/models_v2/alex.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,16 +10,16 @@ class Alex(chainer.Chain):
insize = 227

def __init__(self):
super(Alex, self).__init__(
conv1=L.Convolution2D(None, 96, 11, stride=4),
conv2=L.Convolution2D(None, 256, 5, pad=2),
conv3=L.Convolution2D(None, 384, 3, pad=1),
conv4=L.Convolution2D(None, 384, 3, pad=1),
conv5=L.Convolution2D(None, 256, 3, pad=1),
fc6=L.Linear(None, 4096),
fc7=L.Linear(None, 4096),
fc8=L.Linear(None, 1000),
)
super(Alex, self).__init__()
with self.init_scope():
self.conv1 = L.Convolution2D(None, 96, 11, stride=4)
self.conv2 = L.Convolution2D(None, 256, 5, pad=2)
self.conv3 = L.Convolution2D(None, 384, 3, pad=1)
self.conv4 = L.Convolution2D(None, 384, 3, pad=1)
self.conv5 = L.Convolution2D(None, 256, 3, pad=1)
self.fc6 = L.Linear(None, 4096)
self.fc7 = L.Linear(None, 4096)
self.fc8 = L.Linear(None, 1000)

def __call__(self, x, t):
h = F.max_pooling_2d(F.local_response_normalization(
Expand Down
42 changes: 21 additions & 21 deletions examples/imagenet/models_v2/googlenet.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,29 +8,29 @@ class GoogLeNet(chainer.Chain):
insize = 224

def __init__(self):
super(GoogLeNet, self).__init__(
conv1=L.Convolution2D(None, 64, 7, stride=2, pad=3),
conv2_reduce=L.Convolution2D(None, 64, 1),
conv2=L.Convolution2D(None, 192, 3, stride=1, pad=1),
inc3a=L.Inception(None, 64, 96, 128, 16, 32, 32),
inc3b=L.Inception(None, 128, 128, 192, 32, 96, 64),
inc4a=L.Inception(None, 192, 96, 208, 16, 48, 64),
inc4b=L.Inception(None, 160, 112, 224, 24, 64, 64),
inc4c=L.Inception(None, 128, 128, 256, 24, 64, 64),
inc4d=L.Inception(None, 112, 144, 288, 32, 64, 64),
inc4e=L.Inception(None, 256, 160, 320, 32, 128, 128),
inc5a=L.Inception(None, 256, 160, 320, 32, 128, 128),
inc5b=L.Inception(None, 384, 192, 384, 48, 128, 128),
loss3_fc=L.Linear(None, 1000),
super(GoogLeNet, self).__init__()
with self.init_scope():
self.conv1 = L.Convolution2D(None, 64, 7, stride=2, pad=3)
self.conv2_reduce = L.Convolution2D(None, 64, 1)
self.conv2 = L.Convolution2D(None, 192, 3, stride=1, pad=1)
self.inc3a = L.Inception(None, 64, 96, 128, 16, 32, 32)
self.inc3b = L.Inception(None, 128, 128, 192, 32, 96, 64)
self.inc4a = L.Inception(None, 192, 96, 208, 16, 48, 64)
self.inc4b = L.Inception(None, 160, 112, 224, 24, 64, 64)
self.inc4c = L.Inception(None, 128, 128, 256, 24, 64, 64)
self.inc4d = L.Inception(None, 112, 144, 288, 32, 64, 64)
self.inc4e = L.Inception(None, 256, 160, 320, 32, 128, 128)
self.inc5a = L.Inception(None, 256, 160, 320, 32, 128, 128)
self.inc5b = L.Inception(None, 384, 192, 384, 48, 128, 128)
self.loss3_fc = L.Linear(None, 1000)

loss1_conv=L.Convolution2D(None, 128, 1),
loss1_fc1=L.Linear(None, 1024),
loss1_fc2=L.Linear(None, 1000),
self.loss1_conv = L.Convolution2D(None, 128, 1)
self.loss1_fc1 = L.Linear(None, 1024)
self.loss1_fc2 = L.Linear(None, 1000)

loss2_conv=L.Convolution2D(None, 128, 1),
loss2_fc1=L.Linear(None, 1024),
loss2_fc2=L.Linear(None, 1000)
)
self.loss2_conv = L.Convolution2D(None, 128, 1)
self.loss2_fc1 = L.Linear(None, 1024)
self.loss2_fc2 = L.Linear(None, 1000)

def __call__(self, x, t):
h = F.relu(self.conv1(x))
Expand Down
71 changes: 40 additions & 31 deletions examples/imagenet/models_v2/googlenetbn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,47 +4,56 @@


class GoogLeNetBN(chainer.Chain):
"""New GoogLeNet of BatchNormalization version

(patched for Chainer version 2).
"""
"""New GoogLeNet of BatchNormalization version."""

insize = 224

def __init__(self):
super(GoogLeNetBN, self).__init__(
conv1=L.Convolution2D(None, 64, 7, stride=2, pad=3, nobias=True),
norm1=L.BatchNormalization(64),
conv2=L.Convolution2D(None, 192, 3, pad=1, nobias=True),
norm2=L.BatchNormalization(192),
inc3a=L.InceptionBN(None, 64, 64, 64, 64, 96, 'avg', 32),
inc3b=L.InceptionBN(None, 64, 64, 96, 64, 96, 'avg', 64),
inc3c=L.InceptionBN(None, 0, 128, 160, 64, 96, 'max', stride=2),
inc4a=L.InceptionBN(None, 224, 64, 96, 96, 128, 'avg', 128),
inc4b=L.InceptionBN(None, 192, 96, 128, 96, 128, 'avg', 128),
inc4c=L.InceptionBN(None, 128, 128, 160, 128, 160, 'avg', 128),
inc4d=L.InceptionBN(None, 64, 128, 192, 160, 192, 'avg', 128),
inc4e=L.InceptionBN(None, 0, 128, 192, 192, 256, 'max', stride=2),
inc5a=L.InceptionBN(None, 352, 192, 320, 160, 224, 'avg', 128),
inc5b=L.InceptionBN(None, 352, 192, 320, 192, 224, 'max', 128),
out=L.Linear(None, 1000),
super(GoogLeNetBN, self).__init__()
with self.init_scope():
self.conv1 = L.Convolution2D(
None, 64, 7, stride=2, pad=3, nobias=True)
self.norm1 = L.BatchNormalization(64)
self.conv2 = L.Convolution2D(None, 192, 3, pad=1, nobias=True)
self.norm2 = L.BatchNormalization(192)
self.inc3a = L.InceptionBN(
None, 64, 64, 64, 64, 96, 'avg', 32)
self.inc3b = L.InceptionBN(
None, 64, 64, 96, 64, 96, 'avg', 64)
self.inc3c = L.InceptionBN(
None, 0, 128, 160, 64, 96, 'max', stride=2)
self.inc4a = L.InceptionBN(
None, 224, 64, 96, 96, 128, 'avg', 128)
self.inc4b = L.InceptionBN(
None, 192, 96, 128, 96, 128, 'avg', 128)
self.inc4c = L.InceptionBN(
None, 128, 128, 160, 128, 160, 'avg', 128)
self.inc4d = L.InceptionBN(
None, 64, 128, 192, 160, 192, 'avg', 128)
self.inc4e = L.InceptionBN(
None, 0, 128, 192, 192, 256, 'max', stride=2)
self.inc5a = L.InceptionBN(
None, 352, 192, 320, 160, 224, 'avg', 128)
self.inc5b = L.InceptionBN(
None, 352, 192, 320, 192, 224, 'max', 128)
self.out = L.Linear(None, 1000)

conva=L.Convolution2D(None, 128, 1, nobias=True),
norma=L.BatchNormalization(128),
lina=L.Linear(None, 1024, nobias=True),
norma2=L.BatchNormalization(1024),
outa=L.Linear(None, 1000),
self.conva = L.Convolution2D(None, 128, 1, nobias=True)
self.norma = L.BatchNormalization(128)
self.lina = L.Linear(None, 1024, nobias=True)
self.norma2 = L.BatchNormalization(1024)
self.outa = L.Linear(None, 1000)

convb=L.Convolution2D(None, 128, 1, nobias=True),
normb=L.BatchNormalization(128),
linb=L.Linear(None, 1024, nobias=True),
normb2=L.BatchNormalization(1024),
outb=L.Linear(None, 1000),
)
self.convb = L.Convolution2D(None, 128, 1, nobias=True)
self.normb = L.BatchNormalization(128)
self.linb = L.Linear(None, 1024, nobias=True)
self.normb2 = L.BatchNormalization(1024)
self.outb = L.Linear(None, 1000)

def __call__(self, x, t):
h = F.max_pooling_2d(
F.relu(self.norm1(self.conv1(x))), 3, stride=2, pad=1)
F.relu(self.norm1(self.conv1(x))), 3, stride=2, pad=1)
h = F.max_pooling_2d(
F.relu(self.norm2(self.conv2(h))), 3, stride=2, pad=1)

Expand Down
26 changes: 13 additions & 13 deletions examples/imagenet/models_v2/nin.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,7 +1,6 @@
import math

import chainer
import chainer.functions as F
import chainer.initializers as I
import chainer.links as L


Expand All @@ -12,17 +11,18 @@ class NIN(chainer.Chain):
insize = 227

def __init__(self):
w = math.sqrt(2) # MSRA scaling
super(NIN, self).__init__(
mlpconv1=L.MLPConvolution2D(
None, (96, 96, 96), 11, stride=4, wscale=w),
mlpconv2=L.MLPConvolution2D(
None, (256, 256, 256), 5, pad=2, wscale=w),
mlpconv3=L.MLPConvolution2D(
None, (384, 384, 384), 3, pad=1, wscale=w),
mlpconv4=L.MLPConvolution2D(
None, (1024, 1024, 1000), 3, pad=1, wscale=w),
)
super(NIN, self).__init__()
conv_init = I.HeNormal() # MSRA scaling

with self.init_scope():
self.mlpconv1 = L.MLPConvolution2D(
None, (96, 96, 96), 11, stride=4, conv_init=conv_init)
self.mlpconv2 = L.MLPConvolution2D(
None, (256, 256, 256), 5, pad=2, conv_init=conv_init)
self.mlpconv3 = L.MLPConvolution2D(
None, (384, 384, 384), 3, pad=1, conv_init=conv_init)
self.mlpconv4 = L.MLPConvolution2D(
None, (1024, 1024, 1000), 3, pad=1, conv_init=conv_init)

def __call__(self, x, t):
h = F.max_pooling_2d(F.relu(self.mlpconv1(x)), 3, stride=2)
Expand Down
104 changes: 45 additions & 59 deletions examples/imagenet/models_v2/resnet50.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,25 +10,24 @@
class BottleNeckA(chainer.Chain):

def __init__(self, in_size, ch, out_size, stride=2):
super(BottleNeckA, self).__init__(
conv1=L.Convolution2D(in_size, ch, 1, stride, 0,
initialW=initializers.HeNormal(),
nobias=True),
bn1=L.BatchNormalization(ch),
conv2=L.Convolution2D(ch, ch, 3, 1, 1,
initialW=initializers.HeNormal(),
nobias=True),
bn2=L.BatchNormalization(ch),
conv3=L.Convolution2D(ch, out_size, 1, 1, 0,
initialW=initializers.HeNormal(),
nobias=True),
bn3=L.BatchNormalization(out_size),

conv4=L.Convolution2D(in_size, out_size, 1, stride, 0,
initialW=initializers.HeNormal(),
nobias=True),
bn4=L.BatchNormalization(out_size),
)
super(BottleNeckA, self).__init__()
initialW = initializers.HeNormal()

with self.init_scope():
self.conv1 = L.Convolution2D(
in_size, ch, 1, stride, 0, initialW=initialW, nobias=True)
self.bn1 = L.BatchNormalization(ch)
self.conv2 = L.Convolution2D(
ch, ch, 3, 1, 1, initialW=initialW, nobias=True)
self.bn2 = L.BatchNormalization(ch)
self.conv3 = L.Convolution2D(
ch, out_size, 1, 1, 0, initialW=initialW, nobias=True)
self.bn3 = L.BatchNormalization(out_size)

self.conv4 = L.Convolution2D(
in_size, out_size, 1, stride, 0,
initialW=initialW, nobias=True)
self.bn4 = L.BatchNormalization(out_size)

def __call__(self, x):
h1 = F.relu(self.bn1(self.conv1(x)))
Expand All @@ -42,20 +41,19 @@ def __call__(self, x):
class BottleNeckB(chainer.Chain):

def __init__(self, in_size, ch):
super(BottleNeckB, self).__init__(
conv1=L.Convolution2D(in_size, ch, 1, 1, 0,
initialW=initializers.HeNormal(),
nobias=True),
bn1=L.BatchNormalization(ch),
conv2=L.Convolution2D(ch, ch, 3, 1, 1,
initialW=initializers.HeNormal(),
nobias=True),
bn2=L.BatchNormalization(ch),
conv3=L.Convolution2D(ch, in_size, 1, 1, 0,
initialW=initializers.HeNormal(),
nobias=True),
bn3=L.BatchNormalization(in_size),
)
super(BottleNeckB, self).__init__()
initialW = initializers.HeNormal()

with self.init_scope():
self.conv1 = L.Convolution2D(
in_size, ch, 1, 1, 0, initialW=initialW, nobias=True)
self.bn1 = L.BatchNormalization(ch)
self.conv2 = L.Convolution2D(
ch, ch, 3, 1, 1, initialW=initialW, nobias=True)
self.bn2 = L.BatchNormalization(ch)
self.conv3 = L.Convolution2D(
ch, in_size, 1, 1, 0, initialW=initialW, nobias=True)
self.bn3 = L.BatchNormalization(in_size)

def __call__(self, x):
h = F.relu(self.bn1(self.conv1(x)))
Expand All @@ -65,23 +63,17 @@ def __call__(self, x):
return F.relu(h + x)


class Block(chainer.Chain):
class Block(chainer.ChainList):

def __init__(self, layer, in_size, ch, out_size, stride=2):
super(Block, self).__init__()
links = [('a', BottleNeckA(in_size, ch, out_size, stride))]
self.add_link(BottleNeckA(in_size, ch, out_size, stride))
for i in range(layer - 1):
links += [('b{}'.format(i + 1), BottleNeckB(out_size, ch))]

for l in links:
self.add_link(*l)
self.forward = links
self.add_link(BottleNeckB(out_size, ch))

def __call__(self, x):
for name, _ in sorted(self.forward):
f = getattr(self, name)
for f in self.children():
x = f(x)

return x


Expand All @@ -90,24 +82,18 @@ class ResNet50(chainer.Chain):
insize = 224

def __init__(self):
super(ResNet50, self).__init__(
conv1=L.Convolution2D(3, 64, 7, 2, 3,
initialW=initializers.HeNormal(),
nobias=True),
bn1=L.BatchNormalization(64),
res2=Block(3, 64, 64, 256, 1),
res3=Block(4, 256, 128, 512),
res4=Block(6, 512, 256, 1024),
res5=Block(3, 1024, 512, 2048),
fc=L.Linear(2048, 1000),
)

def clear(self):
self.loss = None
self.accuracy = None
super(ResNet50, self).__init__()
with self.init_scope():
self.conv1 = L.Convolution2D(
3, 64, 7, 2, 3, initialW=initializers.HeNormal())
self.bn1 = L.BatchNormalization(64)
self.res2 = Block(3, 64, 64, 256, 1)
self.res3 = Block(4, 256, 128, 512)
self.res4 = Block(6, 512, 256, 1024)
self.res5 = Block(3, 1024, 512, 2048)
self.fc = L.Linear(2048, 1000)

def __call__(self, x, t):
self.clear()
h = self.bn1(self.conv1(x))
h = F.max_pooling_2d(F.relu(h), 3, stride=2)
h = self.res2(h)
Expand Down