Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/1044197988/TF.Keras-Commonly-used-models
基于Tensorflow的常用模型,包括分类分割、新型激活、卷积模块,可在Tensorflow2.X下运行。
https://github.com/1044197988/TF.Keras-Commonly-used-models
dice-loss image-classification image-segmentation jaccard-loss keras segmentation-loss tensorflow tensorflow2 unet
Last synced: 3 months ago
JSON representation
基于Tensorflow的常用模型,包括分类分割、新型激活、卷积模块,可在Tensorflow2.X下运行。
- Host: GitHub
- URL: https://github.com/1044197988/TF.Keras-Commonly-used-models
- Owner: 1044197988
- License: apache-2.0
- Created: 2019-03-13T12:35:02.000Z (over 5 years ago)
- Default Branch: master
- Last Pushed: 2020-06-27T05:28:02.000Z (over 4 years ago)
- Last Synced: 2024-04-08T02:54:30.050Z (7 months ago)
- Topics: dice-loss, image-classification, image-segmentation, jaccard-loss, keras, segmentation-loss, tensorflow, tensorflow2, unet
- Language: Python
- Homepage:
- Size: 133 KB
- Stars: 173
- Watchers: 7
- Forks: 78
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
- Awesome-Tensorflow2 - 1044197988/TF.Keras-Commonly-used-models
README
# TF.Keras-常用型号
__自己整理的一些tensorflow下ķeras实现的模型,可在Tensorflow2.X下运行__
## 提示:以下模型均不包含预训练权重的载入,只是模型的实现;不同的卷积模块大部分在分类分割模型中已包含。
## 分类模型:
* AlexNet
* Darknet53
* DenseNet
* Dual_path_network
* GoogleNet
* MNasNet
* Resnet34
* Resnet50
* SEResNeXt
* VGG16
* Squeeze_Excite-Network
* MobileNetV3
* Efficientnet
* SE_HRNet
* ResNest## 分割模型:
* FCN8S
* ICNet
* MiniNetv2
* PSPNet-ResNet50
* RAUNet-3D
* Refinenet
* Segnet
* Unet
* Unet_Xception_Resnetblock
* ResNextFPN
* Deeplabv2
* Deeplabv3+
* FastFCN
* HRNet
* ResUNet-a
* RCNN-UNet
* Attention Unet
* RCNN-Attention Unet
* UNet ++
### Unet_family:
#### 不同种类的Unet模型图像分割的实现
1、UNet -U-Net:用于生物医学图像分割的卷积网络 https://arxiv.org/abs/1505.04597
2、RCNN-UNet-基于U-Net的递归残积卷积神经网络(R2U-Net)用于医学图像分割 https://arxiv.org/abs/1802.06955
3、Attention Unet -Attention U-Net:学习在哪里寻找胰腺 https://arxiv.org/abs/1804.03999
4、RCNN-Attention Unet -Attention R2U-Net:只需将两个最新的高级作品集成在一起(R2U-Net + Attention U-Net)
5、嵌套的UNet -UNet ++:用于医学图像分割的嵌套U-Net体系结构 https://arxiv.org/abs/1807.10165
#### 参考:
[Unet-Segmentation-Pytorch-Nest-of-Unets](https://github.com/bigmb/Unet-Segmentation-Pytorch-Nest-of-Unets)
不同点:我的实现初始滤波数32,原始为64。### 分割损失函数:
* Focal_Tversky_loss
* C_Focal_loss
* B_Focal_loss
* LovaszSoftmax
* WeightedCCE
* jaccard_loss
* bce_jaccard_loss
* cce_jaccard_loss
* dice_loss
* bce_dice_loss
* cce_dice_loss### 分割指标:
* iou_score
* jaccard_score
* f1_score
* f2_score
* dice_score### 新型激活函数:
* gelu
* swish
* mish### 卷积模块:
* SE
* Res2Net
* Deformable_Conv### Layer:
* FRN
* attention(PAM空间注意力和CAM通道注意力)
* BiFPN### Others:
* TCN(时间卷积网络——解决LSTM的并发问题)