iFCN_Pytorch
修改torchvision给出的分类模型,实现FCN,并在FCN的基础上,实现iFCN。
相关介绍/实现细节
参考论文
- AlexNet: One weird trick for parallelizing convolutional neural networks
- VGG: Very Deep Convolutional Networks for Large-Scale Image Recognition
- GoogLeNet: Going Deeper with Convolutions
- ResNet: Deep Residual Learning for Image Recognition
- FCN: Fully Convolutional Networks for Semantic Segmentation
- iFCN: Deep Interactive Object Selection
iFCN模型以及验证集评估指标
注:以下模型未经过调参与优化,是未完全收敛的模型。
AlexNet
网络名称 | IoU (voc2012 val) |
NOC (85% IoU) |
epochs | batch size for training |
模型大小 | 模型下载地址 |
---|---|---|---|---|---|---|
AlexNet_32s_deconv | 48.1% | 19.0 | 11 | 128 | 142MB | 下载 |
AlexNet_16s_deconv | 50.5% | 18.8 | 18 | 128 | 78MB | 下载 |
AlexNet_8s_deconv | 54.6% | 17.9 | 29 | 128 | 78MB | 下载 |
VGG系列
网络名称 | IoU (voc2012 val) |
NOC (85% IoU) |
epochs | batch size for training |
模型大小 | 模型下载地址 |
---|---|---|---|---|---|---|
VGG11_32s_deconv | 51.9% | 18.6 | 8 | 48 | 171MB | 下载 |
VGG11_16s_deconv | 52.0% | 19.1 | 15 | 48 | 107MB | 下载 |
VGG11_8s_deconv | 54.6% | 18.7 | 9 | 48 | 107MB | 下载 |
VGG13_32s_deconv | 52.2% | 18.5 | 11 | 32 | 172MB | 下载 |
VGG13_16s_deconv | 52.9% | 18.9 | 11 | 32 | 108MB | 下载 |
VGG13_8s_deconv | 59.4% | 17.9 | 9 | 32 | 108MB | 下载 |
VGG16_32s_deconv | 52.4% | 18.5 | 8 | 32 | 192MB | 下载 |
VGG16_16s_deconv | 56.0% | 18.7 | 7 | 32 | 128MB | 下载 |
VGG16_8s_deconv | 58.2% | 18.2 | 9 | 32 | 128MB | 下载 |
VGG19_32s_deconv | 53.2% | 18.5 | 13 | 32 | 212MB | 下载 |
VGG19_16s_deconv | 56.2% | 18.8 | 17 | 32 | 149MB | 下载 |
VGG19_8s_deconv | 61.7% | 16.7 | 10 | 32 | 149MB | 下载 |
GoogLeNet
网络名称 | IoU (voc2012 val) |
NOC (85% IoU) |
epochs | batch size for training |
模型大小 | 模型下载地址 |
---|---|---|---|---|---|---|
GoogLeNet_32s_deconv | 63.1% | 16.1 | 7 | 48 | 41MB | 下载 |
GoogLeNet_16s_deconv | 66.7% | 14.9 | 15 | 48 | 26MB | 下载 |
GoogLeNet_8s_deconv | 65.3% | 15.7 | 10 | 48 | 26MB | 下载 |
ResNet系列
网络名称 | IoU (voc2012 val) |
NOC (85% IoU) |
epochs | batch size for training |
模型大小 | 模型下载地址 |
---|---|---|---|---|---|---|
ResNet18_32s_deconv | 61.0% | 16.7 | 8 | 96 | 53MB | 下载 |
ResNet18_16s_deconv | 63.8% | 16.5 | 6 | 96 | 45MB | 下载 |
ResNet18_8s_deconv | 70.9% | 13.3 | 11 | 96 | 45MB | 下载 |
ResNet34_32s_deconv | 61.5% | 16.9 | 12 | 64 | 91MB | 下载 |
ResNet34_16s_deconv | 64.2% | 16.5 | 9 | 64 | 83MB | 下载 |
ResNet34_8s_deconv | 72.2% | 12.6 | 22 | 64 | 83MB | 下载 |
ResNet50_32s_deconv | 62.6% | 16.2 | 16 | 32 | 130MB | 下载 |
ResNet50_16s_deconv | 64.8% | 15.2 | 8 | 32 | 98MB | 下载 |
ResNet50_8s_deconv | 71.8% | 12.4 | 31 | 32 | 98MB | 下载 |
ResNet101_32s_deconv | 63.0% | 16.0 | 20 | 24 | 203MB | 下载 |
ResNet101_16s_deconv | 65.4% | 15.2 | 13 | 24 | 171MB | 下载 |
ResNet101_8s_deconv | 73.1% | 12.2 | 17 | 24 | 171MB | 下载 |
ResNet152_32s_deconv | 62.3% | 16.3 | 8 | 16 | 263MB | 下载 |
ResNet152_16s_deconv | 64.7% | 15.5 | 12 | 16 | 231MB | 下载 |
ResNet152_8s_deconv | 73.3% | 11.9 | 13 | 16 | 231MB | 下载 |