Yes it does! To begin with it is identity but it very much has learnable parameters in it.
The skip connections has learnable params when we’re changing number of channels. Also, there’s pooling if stride is anything other than 1.
This is where aman explains convolutions really nicely!
4 Likes
Repeat Blocks - nn.Sequential? combined with nn.ModuleList?
1 Like
The VS Code text size is still a little small for me, might be might resolution though 
Edit: Looks perfect now, Thanks! 
1 Like
channels typo in for loop in cell 7
1 Like
typo in your code *channels
As we build the model, is there a way to display what the connections/network looks like?
channels[stage_idx - 1]?
1 Like
I wanted to know will it be wise to try and use Numpy and Pandas instead of PyTorch for getting to know the intricacies involved?
if prev num_chan <> num_chan … then use prev as from size
1 Like
Would a 1x1 conv layer be useful here as a bridge between 64 and 128,
cool, you got it working.
1 Like
@ramesh says:
Let us complete this as HW and post our Colab Notebooks here for ResNet18, 34 and stretch goal of ResNet50?
I loved when you went whoa! I really felt the happiness 
2 Likes
I too felt the eureka moment !!
2 Likes
bye everyone! good night!
Hi guys,
I was trying to code a resnet50 implementation in pytorch after our reading group.
I could get the model definition and all correct. However when I do a forward pass with batch size of 2 with my custom resnet it works fine, with a batch size of 16 it gives cuda memory out error.
But when I do the same using torchvision's inbuilt resnet50() model, I don’t get this error and the forward pass happens very properly
I tried to check the model parameters for both my custom model and the resnet50 default model. And they seem to be almost the same.
Any ideas on why this discrepency? That is, my forward pass successfully happens with inbuilt resnet vs custom resnet although both have same number of parameters for the same batch size…
Thanks,
Vinayak.
PS: Code for blocks, for the model and the printed model is attached below.
Block Code:
class BottleneckBlock(nn.Module):
def __init__(self, in_channels, out_channels):
super().__init__()
if in_channels == 64: out_channels = 256
if in_channels != out_channels:
interim_channels = in_channels if in_channels == 64 else in_channels // 2
else:
interim_channels = in_channels // 4
self.conv1 = nn.Conv2d(in_channels, interim_channels, kernel_size = 1, bias = False)
self.bn1 = nn.BatchNorm2d(interim_channels)
self.act1 = nn.ReLU(inplace = True)
self.conv2 = nn.Conv2d(interim_channels, interim_channels, kernel_size = 3, padding = 1, bias = False)
self.bn2 = nn.BatchNorm2d(interim_channels)
self.act2 = nn.ReLU(inplace = True)
self.conv3 = nn.Conv2d(interim_channels, out_channels, kernel_size = 1, bias = False)
self.bn3 = nn.BatchNorm2d(out_channels)
self.downsample = noop
self.act3 = nn.ReLU(inplace = True)
if in_channels != out_channels:
self.downsample = nn.Sequential(nn.Conv2d(in_channels, out_channels, kernel_size = 1, bias = False),
nn.BatchNorm2d(out_channels))
def forward(self, x):
original_input = x
# Pass through first conv layer
x = self.act1(self.bn1(self.conv1(x)))
# Pass through the second conv layer
x = self.act2(self.bn2(self.conv2(x)))
# Pass through the third conv layer
x = self.bn3(self.conv3(x))
# Skip Connection
downsample_output = self.downsample(original_input)
# Final output
return self.act3(x + downsample_output)
This is the architecture code
class custom_resnet_bottleneck(nn.Module):
def __init__(self, block_sizes = [3, 4, 6, 3], layers = [64, 256, 512, 1024], classes = 10):
super().__init__()
named_blocks = []
for idx, (bs, ls) in enumerate(zip(block_sizes, layers)):
items = []
for i in range(bs):
if idx == 0:
if i == 0:
bk = BottleneckBlock(ls, layers[idx + 1])
else:
bk = BottleneckBlock(layers[idx + 1], layers[idx + 1])
else:
if i == 0:
bk = BottleneckBlock(ls, 2 * ls)
else:
bk = BottleneckBlock(2 * ls, 2 * ls)
items.append(bk)
named_blocks.append([f"layer_{idx + 1}", nn.Sequential(*items)])
# Define the backbone of the architecture
self.conv1 = nn.Conv2d(3, 64, kernel_size = 7, stride = 2, bias = False, padding = 3)
self.bn1 = nn.BatchNorm2d(64)
self.act1 = nn.ReLU(inplace = True)
self.maxpool = nn.MaxPool2d(kernel_size = 3, stride = 2, padding = 1)
for block in named_blocks:
self.add_module(*block)
# Define the head/classification layer of the architecture
self.head = nn.Sequential(OrderedDict([
('avgpool',nn.AdaptiveAvgPool2d(output_size = (1, 1))),
('flatten', Flatten()),
('fc',nn.Linear(in_features = layers[-1] * 2, out_features = classes)),
]))
def forward(self, x):
x = self.conv1(x)
x = self.bn1(x)
x = self.act1(x)
x = self.maxpool(x)
x = self.layer_1(x)
x = self.layer_2(x)
x = self.layer_3(x)
x = self.layer_4(x)
op = self.head(x)
return op
And this is how the final model looks like
custom_resnet_bottleneck(
(conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act1): ReLU(inplace=True)
(maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
(layer_1): Sequential(
(0): BottleneckBlock(
(conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act1): ReLU(inplace=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act2): ReLU(inplace=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act3): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): BottleneckBlock(
(conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act1): ReLU(inplace=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act2): ReLU(inplace=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act3): ReLU(inplace=True)
)
(2): BottleneckBlock(
(conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act1): ReLU(inplace=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act2): ReLU(inplace=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act3): ReLU(inplace=True)
)
)
(layer_2): Sequential(
(0): BottleneckBlock(
(conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act1): ReLU(inplace=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act2): ReLU(inplace=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act3): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(256, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): BottleneckBlock(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act1): ReLU(inplace=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act2): ReLU(inplace=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act3): ReLU(inplace=True)
)
(2): BottleneckBlock(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act1): ReLU(inplace=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act2): ReLU(inplace=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act3): ReLU(inplace=True)
)
(3): BottleneckBlock(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act1): ReLU(inplace=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act2): ReLU(inplace=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act3): ReLU(inplace=True)
)
)
(layer_3): Sequential(
(0): BottleneckBlock(
(conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act1): ReLU(inplace=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act2): ReLU(inplace=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act3): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): BottleneckBlock(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act1): ReLU(inplace=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act2): ReLU(inplace=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act3): ReLU(inplace=True)
)
(2): BottleneckBlock(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act1): ReLU(inplace=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act2): ReLU(inplace=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act3): ReLU(inplace=True)
)
(3): BottleneckBlock(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act1): ReLU(inplace=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act2): ReLU(inplace=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act3): ReLU(inplace=True)
)
(4): BottleneckBlock(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act1): ReLU(inplace=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act2): ReLU(inplace=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act3): ReLU(inplace=True)
)
(5): BottleneckBlock(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act1): ReLU(inplace=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act2): ReLU(inplace=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act3): ReLU(inplace=True)
)
)
(layer_4): Sequential(
(0): BottleneckBlock(
(conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act1): ReLU(inplace=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act2): ReLU(inplace=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act3): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): BottleneckBlock(
(conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act1): ReLU(inplace=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act2): ReLU(inplace=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act3): ReLU(inplace=True)
)
(2): BottleneckBlock(
(conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act1): ReLU(inplace=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act2): ReLU(inplace=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act3): ReLU(inplace=True)
)
)
(head): Sequential(
(avgpool): AdaptiveAvgPool2d(output_size=(1, 1))
(flatten): Flatten(full=False)
(fc): Linear(in_features=2048, out_features=10, bias=True)
)
)
Hey @vinayak_nayak - easiest to just share a colab notebook that replicates this error. I’ll take a look - thanks! 
1 Like
Our own implementation is using more memory than it should - could be a for loop somewhere, an incorrect if statement or something else really!
1 Like
I will Aman, thanks a lot!
1 Like