def __init__(self,
num_classes=10,
num_channels=3,
dimensions=(28, 28),
fc_inputs=12,
bias=False):
super().__init__()
# AI84 Limits
assert dimensions[0] == dimensions[1] # Only square supported
# Keep track of image dimensions so one constructor works for all image sizes
dim = dimensions[0]
self.conv1 = ai8x.FusedConv2dReLU(num_channels,
16,
3,
padding=1,
bias=bias)
# padding 1 -> no change in dimensions -> 16x28x28
pad = 2 if dim == 28 else 1
self.conv2 = ai8x.FusedMaxPoolConv2dReLU(16,
16,
3,
pool_size=2,
pool_stride=2,
padding=pad,
bias=bias)
dim //= 2 # pooling, padding 0 -> 16x14x14
if pad == 2:
dim += 2 # padding 2 -> 16x16x16
self.conv3 = ai8x.FusedMaxPoolConv2dReLU(16,
fc_inputs,
3,
pool_size=4,
pool_stride=4,
padding=1,
bias=bias)
dim //= 4 # pooling, padding 0 -> 16x4x4
# padding 1 -> 12x4x4
self.fc = ai8x.SoftwareLinear(fc_inputs * dim * dim,
num_classes,
bias=True)
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight,
mode='fan_out',
nonlinearity='relu')
def __init__(self,
num_channels=3,
num_classes=10,
dimensions=(32, 32),
bias=False,
**kwargs):
super().__init__()
dim1 = dimensions[0]
dim2 = dimensions[1]
# 3x32x32
self.conv1 = ai8x.FusedMaxPoolConv2dReLU(in_channels=num_channels,
out_channels=64,
kernel_size=3,
padding=1,
bias=bias,
**kwargs)
dim1 //= 2
dim2 //= 2
# 64x16x16
self.fire1 = ai8x_fire.Fire(in_planes=64,
squeeze_planes=16,
expand1x1_planes=64,
expand3x3_planes=64,
bias=bias,
**kwargs)
# 128x16x16
self.fire2 = ai8x_fire.Fire(in_planes=128,
squeeze_planes=16,
expand1x1_planes=64,
expand3x3_planes=64,
bias=bias,
**kwargs)
# 128x16x16
self.pool1 = nn.MaxPool2d(kernel_size=2, stride=2,
padding=0) # check if kernel size=3
dim1 //= 2
dim2 //= 2
# 128x8x8
self.fire3 = ai8x_fire.Fire(in_planes=128,
squeeze_planes=32,
expand1x1_planes=128,
expand3x3_planes=128,
bias=bias,
**kwargs)
# 256x8x8
self.fire4 = ai8x_fire.Fire(in_planes=256,
squeeze_planes=32,
expand1x1_planes=128,
expand3x3_planes=128,
bias=bias,
**kwargs)
# 256x8x8
self.pool2 = nn.MaxPool2d(kernel_size=2, stride=2,
padding=0) # check if kernel size=3
dim1 //= 2
dim2 //= 2
# 256x4x4
self.fire5 = ai8x_fire.Fire(in_planes=256,
squeeze_planes=48,
expand1x1_planes=192,
expand3x3_planes=192,
bias=bias,
**kwargs)
# 384x4x4
self.fire6 = ai8x_fire.Fire(in_planes=384,
squeeze_planes=48,
expand1x1_planes=192,
expand3x3_planes=192,
bias=bias,
**kwargs)
# 384x4x4
self.fire7 = ai8x_fire.Fire(in_planes=384,
squeeze_planes=48,
expand1x1_planes=256,
expand3x3_planes=256,
bias=bias,
**kwargs)
# 512x4x4
self.fire8 = ai8x_fire.Fire(in_planes=512,
squeeze_planes=64,
expand1x1_planes=256,
expand3x3_planes=256,
bias=bias,
**kwargs)
# 512x4x4
# self.conv2 = ai8x.FusedAvgPoolConv2dReLU(in_channels=512, out_channels=num_classes,
# kernel_size=1, pool_size=4, pool_stride=4)
self.fc = ai8x.SoftwareLinear(512 * dim1 * dim2,
num_classes,
bias=bias)
# num_classesx1x1
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight,
mode='fan_out',
nonlinearity='relu')
def __init__(self,
num_classes=10,
num_channels=3,
dimensions=(28, 28),
planes=128,
pool=2,
fc_inputs=12,
bias=False,
**kwargs):
super().__init__()
# Keep track of image dimensions so one constructor works for all image sizes
dim = dimensions[0]
self.conv1 = ai8x.FusedConv2dReLU(num_channels,
planes,
3,
padding=1,
bias=bias,
**kwargs)
# padding 1 -> no change in dimensions -> MNIST: 28x28 | CIFAR: 32x32
pad = 2 if dim == 28 else 1
self.conv2 = ai8x.FusedMaxPoolConv2dReLU(planes,
60,
3,
pool_size=2,
pool_stride=2,
padding=pad,
bias=bias,
**kwargs)
dim //= 2 # pooling, padding 0 -> MNIST: 14x14 | CIFAR: 16x16
if pad == 2:
dim += 2 # MNIST: padding 2 -> 16x16 | CIFAR: padding 1 -> 16x16
self.conv3 = ai8x.FusedMaxPoolConv2dReLU(60,
56,
3,
pool_size=2,
pool_stride=2,
padding=1,
bias=bias,
**kwargs)
dim //= 2 # pooling, padding 0 -> 8x8
# padding 1 -> no change in dimensions
self.conv4 = ai8x.FusedAvgPoolConv2dReLU(56,
fc_inputs,
3,
pool_size=pool,
pool_stride=2,
padding=1,
bias=bias,
**kwargs)
dim //= pool # pooling, padding 0 -> 4x4
# padding 1 -> no change in dimensions
self.fc = ai8x.SoftwareLinear(fc_inputs * dim * dim,
num_classes,
bias=True)
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight,
mode='fan_out',
nonlinearity='relu')
