代码拉取完成，页面将自动刷新
This repository contains code for Deformable ConvNets v2 (Modulated Deformable Convolution) based on Deformable ConvNets v2: More Deformable, Better Results implemented in PyTorch. This implementation of deformable convolution based on ChunhuanLin/deform_conv_pytorch, thanks to ChunhuanLin.
Replace regular convolution (following model's conv2) with modulated deformable convolution:
class ConvNet(nn.Module):
def __init__(self):
self.relu = nn.ReLU(inplace=True)
self.pool = nn.MaxPool2d((2, 2))
self.avg_pool = nn.AdaptiveAvgPool2d(1)
self.conv1 = nn.Conv2d(3, 32, 3, padding=1)
self.bn1 = nn.BatchNorm2d(32)
self.conv2 = nn.DeformConv2d(32, 64, 3, padding=1, modulation=True)
self.bn2 = nn.BatchNorm2d(64)
self.fc = nn.Linear(64, 10)
def forward(self, x):
x = self.relu(self.bn1(self.conv1(x)))
x = self.pool(x)
x = self.relu(self.bn2(self.conv2(x)))
x = self.avg_pool(x)
x = x.view(x.shape[0], -1)
x = self.fc(x)
return x
ScaledMNIST is randomly scaled MNIST.
Use modulated deformable convolution at conv3~4:
python train.py --arch ScaledMNISTNet --deform True --modulation True --min-deform-layer 3
Use deformable convolution at conv3~4:
python train.py --arch ScaledMNISTNet --deform True --modulation False --min-deform-layer 3
Use only regular convolution:
python train.py --arch ScaledMNISTNet --deform False --modulation False
| Model | Accuracy (%) | Loss |
|---|---|---|
| w/o DCN | 97.22 | 0.113 |
| w/ DCN @conv4 | 98.60 | 0.049 |
| w/ DCN @conv3~4 | 98.95 | 0.035 |
| w/ DCNv2 @conv4 | 98.45 | 0.058 |
| w/ DCNv2 @conv3~4 | 99.21 | 0.027 |
此处可能存在不合适展示的内容，页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容，可点击提交进行申诉，我们将尽快为您处理。
马建仓 AI 助手
