提交 f10df497 编写于 作者: C cuicheng01 提交者: ruri

update some pretrained models in image classification (#3315)

refine image classification and add 19 pre-trained models.
上级 22c785db
......@@ -136,7 +136,7 @@ bash run.sh train 模型名
* **use_gpu**: 是否在GPU上运行,默认值: True
* **use_label_smoothing**: 是否对数据进行label smoothing处理,默认值: False
* **label_smoothing_epsilon**: label_smoothing的epsilon, 默认值:0.2
* **label_smoothing_epsilon**: label_smoothing的epsilon, 默认值:0.1
* **random_seed**: 随机数种子, 默认值: 1000
**数据读取器说明:** 数据读取器定义在```reader.py```文件中,现在默认基于cv2的数据读取器, 在[训练阶段](#模型训练),默认采用的增广方式是随机裁剪与水平翻转, 而在[模型评估](#模型评估)[模型预测](#模型预测)阶段用的默认方式是中心裁剪。当前支持的数据增广方式有:
......@@ -255,12 +255,17 @@ PaddlePaddle/Models ImageClassification 支持自定义数据
### MobileNet Series
|Model | Top-1 | Top-5 | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |:-: |
|[MobileNetV1_x0_25](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV1_x0_25_pretrained.tar) | 51.43% | 75.46% | 2.283 | 0.866 |
|[MobileNetV1_x0_5](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV1_x0_5_pretrained.tar) | 63.52% | 84.73% | 2.378 | 1.058 |
|[MobileNetV1_x0_75](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV1_x0_75_pretrained.tar) | 68.81% | 88.23% | 2.540 | 1.386 |
|[MobileNetV1](http://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV1_pretrained.tar) | 70.99% | 89.68% | 2.609 |1.615 |
|[MobileNetV2](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV2_pretrained.tar) | 72.15% | 90.65% | 4.546 | 5.278 |
|[MobileNetV2_x0_25](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV2_x0_25_pretrained.tar) | 53.21% | 76.52% | 4.267 | 3.777 |
|[MobileNetV2_x0_5](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV2_x0_5_pretrained.tar) | 65.03% | 85.72% | 4.514 | 4.150 |
|[MobileNetV2_x0_75](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV2_x0_75_pretrained.tar) | 69.83% | 89.01% | 4.313 | 3.720 |
|[MobileNetV2](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV2_pretrained.tar) | 72.15% | 90.65% | 4.546 | 5.278 |
|[MobileNetV2_x1_5](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV2_x1_5_pretrained.tar) | 74.12% | 91.67% | 5.235 | 6.909 |
|[MobileNetV2_x2_0](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV2_x2_0_pretrained.tar) | 75.23% | 92.58% | 6.680 | 7.658 |
|[MobileNetV3_small_x1_0](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV3_small_x1_0_pretrained.tar) | 67.46% | 87.12% | 6.809 | |
### ShuffleNet Series
|Model | Top-1 | Top-5 | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
......@@ -277,7 +282,9 @@ PaddlePaddle/Models ImageClassification 支持自定义数据
|Model | Top-1 | Top-5 | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |:-: |
|[ResNet18](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet18_pretrained.tar) | 70.98% | 89.92% | 3.456 | 2.484 |
|[ResNet18_vd](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet18_vd_pretrained.tar) | 72.26% | 90.80% | 3.847 | 2.473 |
|[ResNet34](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet34_pretrained.tar) | 74.57% | 92.14% | 5.668 | 3.767 |
|[ResNet34_vd](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet34_vd_pretrained.tar) | 75.98% | 92.98% | 6.089 | 3.531 |
|[ResNet50](http://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_pretrained.tar) | 76.50% | 93.00% | 8.787 | 5.434 |
|[ResNet50_vc](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_vc_pretrained.tar) |78.35% | 94.03% | 9.013 | 5.463 |
|[ResNet50_vd](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_vd_pretrained.tar) | 79.12% | 94.44% | 9.058 | 5.510 |
......@@ -296,10 +303,12 @@ PaddlePaddle/Models ImageClassification 支持自定义数据
|[ResNeXt50_64x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt50_64x4d_pretrained.tar) | 78.43% | 94.13% | 28.162 | 18.271 |
|[ResNeXt50_vd_64x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt50_vd_64x4d_pretrained.tar) | 80.12% | 94.86% | 20.888 | 17.687 |
|[ResNeXt101_32x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt101_32x4d_pretrained.tar) | 78.65% | 94.19% | 24.154 | 21.387 |
|[ResNeXt101_vd_32x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt101_vd_32x4d_pretrained.tar) | 80.33% | 95.12% | 24.701 | 18.032 |
|[ResNeXt101_64x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt50_64x4d_pretrained.tar) | 78.43% | 94.13% | 41.073 | 38.736 |
|[ResNeXt101_vd_64x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt101_vd_64x4d_pretrained.tar) | 80.78% | 95.20% | 42.277 | 40.929 |
|[ResNeXt152_32x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt152_32x4d_pretrained.tar) | 78.98% | 94.33% | 37.007 | 31.301 |
|[ResNeXt152_64x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt152_64x4d_pretrained.tar) | 79.51% | 94.71% | 58.966 | 57.267 |
|[ResNeXt152_vd_64x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt152_vd_64x4d_pretrained.tar) | 81.08% | 95.34% | 60.947 | 49.117 |
### DenseNet Series
|Model | Top-1 | Top-5 | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
......@@ -310,9 +319,19 @@ PaddlePaddle/Models ImageClassification 支持自定义数据
|[DenseNet201](https://paddle-imagenet-models-name.bj.bcebos.com/DenseNet201_pretrained.tar) | 77.63% | 93.66% | 26.583 | 10.549 |
|[DenseNet264](https://paddle-imagenet-models-name.bj.bcebos.com/DenseNet264_pretrained.tar) | 77.96% | 93.85% | 41.495 | 15.574 |
### DPN Series
|Model | Top-1 | Top-5 | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |:-: |
|[DPN68](https://paddle-imagenet-models-name.bj.bcebos.com/DPN68_pretrained.tar) | 76.78% | 93.43% | 18.446 | 6.324 |
|[DPN92](https://paddle-imagenet-models-name.bj.bcebos.com/DPN92_pretrained.tar) | 79.85% | 94.80% | 25.748 | 22.182 |
|[DPN98](https://paddle-imagenet-models-name.bj.bcebos.com/DPN98_pretrained.tar) | 80.59% | 95.10% | 29.421 | 13.657 |
|[DPN107](https://paddle-imagenet-models-name.bj.bcebos.com/DPN107_pretrained.tar) | 80.89% | 95.32% | 41.071 | 19.115 |
|[DPN131](https://paddle-imagenet-models-name.bj.bcebos.com/DPN131_pretrained.tar) | 80.70% | 95.14% | 41.179 | 18.278 |
### SENet Series
|Model | Top-1 | Top-5 | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |:-: |
|[SE_ResNet50_vd](https://paddle-imagenet-models-name.bj.bcebos.com/SE_ResNet50_vd_pretrained.tar) | 79.52% | 94.75% | 10.345 | 7.662 |
|[SE_ResNeXt50_32x4d](https://paddle-imagenet-models-name.bj.bcebos.com/SE_ResNeXt50_32x4d_pretrained.tar) | 78.44% | 93.96% | 14.916 | 12.126 |
|[SE_ResNeXt101_32x4d](https://paddle-imagenet-models-name.bj.bcebos.com/SE_ResNeXt101_32x4d_pretrained.tar) | 79.12% | 94.20% | 30.085 | 24.110 |
|[SENet_154_vd](https://paddle-imagenet-models-name.bj.bcebos.com/SENet_154_vd_pretrained.tar) | 81.40% | 95.48% | 71.892 | 64.855 |
......@@ -321,7 +340,11 @@ PaddlePaddle/Models ImageClassification 支持自定义数据
| Model | Top-1 | Top-5 | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |:-: |
|[GoogLeNet](https://paddle-imagenet-models-name.bj.bcebos.com/GoogLeNet_pretrained.tar) | 70.70% | 89.66% | 6.528 | 3.076 |
|[Xception_41](https://paddle-imagenet-models-name.bj.bcebos.com/Xception_41_pretrained.tar) | 79.30% | 94.53% | 13.757 | 10.831 |
|[Xception41](https://paddle-imagenet-models-name.bj.bcebos.com/Xception41_pretrained.tar) | 79.30% | 94.53% | 13.757 | 10.831 |
|[Xception41_deeplab](https://paddle-imagenet-models-name.bj.bcebos.com/Xception41_deeplab_pretrained.tar) | 79.55% | 94.38% | 14.268 | 10.301 |
|[Xception65](https://paddle-imagenet-models-name.bj.bcebos.com/Xception65_pretrained.tar) | 81.00% | 95.49% | 19.216 | 15.981 |
|[Xception65_deeplab](https://paddle-imagenet-models-name.bj.bcebos.com/Xception65_deeplab_pretrained.tar) | 80.32% | 94.49% | 19.536 | 16.365 |
|[Xception71](https://paddle-imagenet-models-name.bj.bcebos.com/Xception71_pretrained.tar) | 81.11% | 95.45% | 23.291 | 18.974 |
|[InceptionV4](https://paddle-imagenet-models-name.bj.bcebos.com/InceptionV4_pretrained.tar) | 80.77% | 95.26% | 32.413 | 18.154 |
### DarkNet
......@@ -366,12 +389,14 @@ PaddlePaddle/Models ImageClassification 支持自定义数据
- ShuffleNetV2: [ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design](https://arxiv.org/abs/1807.11164), Ningning Ma, Xiangyu Zhang, Hai-Tao Zheng, Jian Sun
- MobileNetV1: [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://arxiv.org/abs/1704.04861), Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, Hartwig Adam
- MobileNetV2: [MobileNetV2: Inverted Residuals and Linear Bottlenecks](https://arxiv.org/pdf/1801.04381v4.pdf), Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, Liang-Chieh Chen
- MobileNetV3: [Searching for MobileNetV3](https://arxiv.org/pdf/1905.02244.pdf), Andrew Howard, Mark Sandler, Grace Chu, Liang-Chieh Chen, Bo Chen, Mingxing Tan, Weijun Wang, Yukun Zhu, Ruoming Pang, Vijay Vasudevan, Quoc V. Le, Hartwig Adam
- VGG: [Very Deep Convolutional Networks for Large-scale Image Recognition](https://arxiv.org/pdf/1409.1556), Karen Simonyan, Andrew Zisserman
- GoogLeNet: [Going Deeper with Convolutions](https://www.cs.unc.edu/~wliu/papers/GoogLeNet.pdf), Christian Szegedy1, Wei Liu2, Yangqing Jia
- Xception: [Xception: Deep Learning with Depthwise Separable Convolutions](https://arxiv.org/abs/1610.02357), Franc ̧ois Chollet
- InceptionV4: [Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning](https://arxiv.org/abs/1602.07261), Christian Szegedy, Sergey Ioffe, Vincent Vanhoucke, Alex Alemi
- DarkNet: [YOLOv3: An Incremental Improvement](https://pjreddie.com/media/files/papers/YOLOv3.pdf), Joseph Redmon, Ali Farhadi
- DenseNet: [Densely Connected Convolutional Networks](https://arxiv.org/abs/1608.06993), Gao Huang, Zhuang Liu, Laurens van der Maaten
- DPN: [Dual Path Networks](https://arxiv.org/pdf/1707.01629.pdf), Yunpeng Chen, Jianan Li, Huaxin Xiao, Xiaojie Jin, Shuicheng Yan, Jiashi Feng
- SqueezeNet: [SQUEEZENET: ALEXNET-LEVEL ACCURACY WITH 50X FEWER PARAMETERS AND <0.5MB MODEL SIZE](https://arxiv.org/abs/1602.07360), Forrest N. Iandola, Song Han, Matthew W. Moskewicz, Khalid Ashraf, William J. Dally, Kurt Keutzer
- ResNeXt101_wsl: [Exploring the Limits of Weakly Supervised Pretraining](https://arxiv.org/abs/1805.00932), Dhruv Mahajan, Ross Girshick, Vignesh Ramanathan, Kaiming He, Manohar Paluri, Yixuan Li, Ashwin Bharambe, Laurens van der Maaten
- Fix_ResNeXt101_wsl: [Fixing the train-test resolution discrepancy](https://arxiv.org/abs/1906.06423), Hugo Touvron, Andrea Vedaldi, Matthijs Douze, Herve ́ Je ́gou
......@@ -383,9 +408,10 @@ PaddlePaddle/Models ImageClassification 支持自定义数据
- 2019/04/01 **Stage3**: 更新ResNet18,ResNet34,GoogLeNet,ShuffleNetV2
- 2019/06/12 **Stage4**: 更新ResNet50_vc,ResNet50_vd,ResNet101_vd,ResNet152_vd,ResNet200_vd,SE154_vd InceptionV4,ResNeXt101_64x4d,ResNeXt101_vd_64x4d
- 2019/06/22 更新ResNet50_vd_v2
- 2019/07/02 **Stage5**: 更新MobileNetV2_x0_5,ResNeXt50_32x4d,ResNeXt50_64x4d,Xception_41,ResNet101_vd
- 2019/07/02 **Stage5**: 更新MobileNetV2_x0_5,ResNeXt50_32x4d,ResNeXt50_64x4d,Xception41,ResNet101_vd
- 2019/07/19 **Stage6**: 更新ShuffleNetV2_x0_25,ShuffleNetV2_x0_33,ShuffleNetV2_x0_5,ShuffleNetV2_x1_0,ShuffleNetV2_x1_5,ShuffleNetV2_x2_0,MobileNetV2_x0_25,MobileNetV2_x1_5,MobileNetV2_x2_0,ResNeXt50_vd_64x4d,ResNeXt101_32x4d,ResNeXt152_32x4d
- 2019/08/01 **Stage7**: 更新DarkNet53,DenseNet121,Densenet161,DenseNet169,DenseNet201,DenseNet264,SqueezeNet1_0,SqueezeNet1_1,ResNeXt50_vd_32x4d,ResNeXt152_64x4d,ResNeXt101_32x8d_wsl,ResNeXt101_32x16d_wsl,ResNeXt101_32x32d_wsl,ResNeXt101_32x48d_wsl,Fix_ResNeXt101_32x48d_wsl
- 2019/09/11 **Stage8**: 更新ResNet18_vd,ResNet34_vd,MobileNetV1_x0_25,MobileNetV1_x0_5,MobileNetV1_x0_75,MobileNetV2_x0_75,MobilenNetV3_small_x1_0,DPN68,DPN92,DPN98,DPN107,DPN131,ResNeXt101_vd_32x4d,ResNeXt152_vd_64x4d,Xception65,Xception71,Xception41_deeplab,Xception65_deeplab,SE_ResNet50_vd
## 如何贡献代码
......
......@@ -128,7 +128,7 @@ Switch:
* **use_gpu**: whether to use GPU or not. Default: True.
* **use_label_smoothing**: whether to use label_smoothing or not. Default:False.
* **label_smoothing_epsilon**: the label_smoothing_epsilon. Default:0.2.
* **label_smoothing_epsilon**: the label_smoothing_epsilon. Default:0.1.
* **random_seed**: random seed for debugging, Default: 1000
**data reader introduction:** Data reader is defined in ```reader.py```, default reader is implemented by opencv. In the [Training](#training) Stage, random crop and flipping are applied, while center crop is applied in the [Evaluation](#evaluation) and [Inference](#inference) stages. Supported data augmentation includes:
......@@ -215,110 +215,133 @@ Pretrained models can be downloaded by clicking related model names.
```
- 4: The pretrained model of the ResNeXt101_wsl series network is converted from the pytorch model. Please refer to [RESNEXT WSL](https://pytorch.org/hub/facebookresearch_WSL-Images_resnext/) for details.
### AlexNet Series
|model | top-1/top-5 accuracy(CV2) | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |
|[AlexNet](http://paddle-imagenet-models-name.bj.bcebos.com/AlexNet_pretrained.tar) | 56.72%/79.17% | 3.083 | 2.728 |
### AlexNet
|Model | Top-1 | Top-5 | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |:-: |
|[AlexNet](http://paddle-imagenet-models-name.bj.bcebos.com/AlexNet_pretrained.tar) | 56.72% | 79.17% | 3.083 | 2.728 |
### SqueezeNet Series
|model | top-1/top-5 accuracy(CV2) | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |
|[SqueezeNet1_0](https://paddle-imagenet-models-name.bj.bcebos.com/SqueezeNet1_0_pretrained.tar) | 59.60%/81.66% | 2.740 | 1.688 |
|[SqueezeNet1_1](https://paddle-imagenet-models-name.bj.bcebos.com/SqueezeNet1_1_pretrained.tar) | 60.08%/81.85% | 2.751 | 1.270 |
### SqueezeNet
|Model | Top-1 | Top-5 | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |:-: |
|[SqueezeNet1_0](https://paddle-imagenet-models-name.bj.bcebos.com/SqueezeNet1_0_pretrained.tar) | 59.60% | 81.66% | 2.740 | 1.688 |
|[SqueezeNet1_1](https://paddle-imagenet-models-name.bj.bcebos.com/SqueezeNet1_1_pretrained.tar) | 60.08% | 81.85% | 2.751 | 1.270 |
### VGG Series
|model | top-1/top-5 accuracy(CV2) | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |
|[VGG11](https://paddle-imagenet-models-name.bj.bcebos.com/VGG11_pretrained.tar) | 69.28%/89.09% | 8.223 | 6.821 |
|[VGG13](https://paddle-imagenet-models-name.bj.bcebos.com/VGG13_pretrained.tar) | 70.02%/89.42% | 9.512 | 7.783 |
|[VGG16](https://paddle-imagenet-models-name.bj.bcebos.com/VGG16_pretrained.tar) | 72.00%/90.69% | 11.315 | 9.067 |
|[VGG19](https://paddle-imagenet-models-name.bj.bcebos.com/VGG19_pretrained.tar) | 72.56%/90.93% | 13.096 | 10.388 |
|Model | Top-1 | Top-5 | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |:-: |
|[VGG11](https://paddle-imagenet-models-name.bj.bcebos.com/VGG11_pretrained.tar) | 69.28% | 89.09% | 8.223 | 6.821 |
|[VGG13](https://paddle-imagenet-models-name.bj.bcebos.com/VGG13_pretrained.tar) | 70.02% | 89.42% | 9.512 | 7.783 |
|[VGG16](https://paddle-imagenet-models-name.bj.bcebos.com/VGG16_pretrained.tar) | 72.00% | 90.69% | 11.315 | 9.067 |
|[VGG19](https://paddle-imagenet-models-name.bj.bcebos.com/VGG19_pretrained.tar) | 72.56% | 90.93% | 13.096 | 10.388 |
### MobileNet Series
|model | top-1/top-5 accuracy(CV2) | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |
|[MobileNetV1](http://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV1_pretrained.tar) | 70.99%/89.68% | 2.609 |1.615 |
|[MobileNetV2_x0_25](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV2_x0_25_pretrained.tar) | 53.21%/76.52% | 4.267 | 3.777 |
|[MobileNetV2_x0_5](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV2_x0_5_pretrained.tar) | 65.03%/85.72% | 4.514 | 4.150 |
|[MobileNetV2_x1_0](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV2_pretrained.tar) | 72.15%/90.65% | 4.546 | 5.278 |
|[MobileNetV2_x1_5](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV2_x1_5_pretrained.tar) | 74.12%/91.67% | 5.235 | 6.909 |
|[MobileNetV2_x2_0](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV2_x2_0_pretrained.tar) | 75.23%/92.58% | 6.680 | 7.658 |
|Model | Top-1 | Top-5 | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |:-: |
|[MobileNetV1_x0_25](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV1_x0_25_pretrained.tar) | 51.43% | 75.46% | 2.283 | 0.866 |
|[MobileNetV1_x0_5](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV1_x0_5_pretrained.tar) | 63.52% | 84.73% | 2.378 | 1.058 |
|[MobileNetV1_x0_75](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV1_x0_75_pretrained.tar) | 68.81% | 88.23% | 2.540 | 1.386 |
|[MobileNetV1](http://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV1_pretrained.tar) | 70.99% | 89.68% | 2.609 |1.615 |
|[MobileNetV2_x0_25](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV2_x0_25_pretrained.tar) | 53.21% | 76.52% | 4.267 | 3.777 |
|[MobileNetV2_x0_5](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV2_x0_5_pretrained.tar) | 65.03% | 85.72% | 4.514 | 4.150 |
|[MobileNetV2_x0_75](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV2_x0_75_pretrained.tar) | 69.83% | 89.01% | 4.313 | 3.720 |
|[MobileNetV2](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV2_pretrained.tar) | 72.15% | 90.65% | 4.546 | 5.278 |
|[MobileNetV2_x1_5](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV2_x1_5_pretrained.tar) | 74.12% | 91.67% | 5.235 | 6.909 |
|[MobileNetV2_x2_0](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV2_x2_0_pretrained.tar) | 75.23% | 92.58% | 6.680 | 7.658 |
|[MobileNetV3_small_x1_0](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV3_small_x1_0_pretrained.tar) | 67.46% | 87.12% | 6.809 | |
### ShuffleNet Series
|model | top-1/top-5 accuracy(CV2) | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |
|[ShuffleNetV2_x0_25](https://paddle-imagenet-models-name.bj.bcebos.com/ShuffleNetV2_x0_25_pretrained.tar) | 49.90%/73.79% | 5.956 | 2.961 |
|[ShuffleNetV2_x0_33](https://paddle-imagenet-models-name.bj.bcebos.com/ShuffleNetV2_x0_33_pretrained.tar) | 53.73%/77.05% | 5.896 | 2.941 |
|[ShuffleNetV2_x0_5](https://paddle-imagenet-models-name.bj.bcebos.com/ShuffleNetV2_x0_5_pretrained.tar) | 60.32%/82.26% | 6.048 | 3.088 |
|[ShuffleNetV2_x1_0](https://paddle-imagenet-models-name.bj.bcebos.com/ShuffleNetV2_x1_0_pretrained.tar) | 68.80%/88.45% | 6.101 | 3.616 |
|[ShuffleNetV2_x1_5](https://paddle-imagenet-models-name.bj.bcebos.com/ShuffleNetV2_x1_5_pretrained.tar) | 71.63%/90.15% | 6.113 | 3.699 |
|[ShuffleNetV2_x2_0](https://paddle-imagenet-models-name.bj.bcebos.com/ShuffleNetV2_x2_0_pretrained.tar) | 73.15%/91.20% | 6.430 | 4.553 |
|[ShuffleNetV2_x1_0_swish](https://paddle-imagenet-models-name.bj.bcebos.com/ShuffleNetV2_pretrained.tar) | 70.03%/89.17% | 6.078 | 6.282 |
|Model | Top-1 | Top-5 | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |:-: |
|[ShuffleNetV2](https://paddle-imagenet-models-name.bj.bcebos.com/ShuffleNetV2_pretrained.tar) | 68.80% | 88.45% | 6.101 | 3.616 |
|[ShuffleNetV2_x0_25](https://paddle-imagenet-models-name.bj.bcebos.com/ShuffleNetV2_x0_25_pretrained.tar) | 49.90% | 73.79% | 5.956 | 2.961 |
|[ShuffleNetV2_x0_33](https://paddle-imagenet-models-name.bj.bcebos.com/ShuffleNetV2_x0_33_pretrained.tar) | 53.73% | 77.05% | 5.896 | 2.941 |
|[ShuffleNetV2_x0_5](https://paddle-imagenet-models-name.bj.bcebos.com/ShuffleNetV2_x0_5_pretrained.tar) | 60.32% | 82.26% | 6.048 | 3.088 |
|[ShuffleNetV2_x1_5](https://paddle-imagenet-models-name.bj.bcebos.com/ShuffleNetV2_x1_5_pretrained.tar) | 71.63% | 90.15% | 6.113 | 3.699 |
|[ShuffleNetV2_x2_0](https://paddle-imagenet-models-name.bj.bcebos.com/ShuffleNetV2_x2_0_pretrained.tar) | 73.15% | 91.20% | 6.430 | 4.553 |
|[ShuffleNetV2_swish](https://paddle-imagenet-models-name.bj.bcebos.com/ShuffleNetV2_swish_pretrained.tar) | 70.03% | 89.17% | 6.078 | 6.282 |
### ResNet Series
|model | top-1/top-5 accuracy(CV2) | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |
|[ResNet18](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet18_pretrained.tar) | 70.98%/89.92% | 3.456 | 2.484 |
|[ResNet34](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet34_pretrained.tar) | 74.57%/92.14% | 5.668 | 3.767 |
|[ResNet50](http://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_pretrained.tar) | 76.50%/93.00% | 8.787 | 5.434 |
|[ResNet50_vc](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_vc_pretrained.tar) |78.35%/94.03% | 9.013 | 5.463 |
|[ResNet50_vd](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_vd_pretrained.tar) | 79.12%/94.44% | 9.058 | 5.510 |
|[ResNet50_vd_v2](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_vd_v2_pretrained.tar) | 79.84%/94.93% | 9.058 | 5.510 |
|[ResNet101](http://paddle-imagenet-models-name.bj.bcebos.com/ResNet101_pretrained.tar) | 77.56%/93.64% | 15.447 | 8.779 |
|[ResNet101_vd](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet101_vd_pretrained.tar) | 80.17%/94.97% | 15.685 | 8.878 |
|[ResNet152](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet152_pretrained.tar) | 78.26%/93.96% | 21.816 | 12.148 |
|[ResNet152_vd](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet152_vd_pretrained.tar) | 80.59%/95.30% | 22.041 | 12.259 |
|[ResNet200_vd](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet200_vd_pretrained.tar) | 80.93%/95.33% | 28.015 | 15.278 |
|Model | Top-1 | Top-5 | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |:-: |
|[ResNet18](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet18_pretrained.tar) | 70.98% | 89.92% | 3.456 | 2.484 |
|[ResNet18_vd](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet18_vd_pretrained.tar) | 72.26% | 90.80% | 3.847 | 2.473 |
|[ResNet34](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet34_pretrained.tar) | 74.57% | 92.14% | 5.668 | 3.767 |
|[ResNet34_vd](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet34_vd_pretrained.tar) | 75.98% | 92.98% | 6.089 | 3.531 |
|[ResNet50](http://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_pretrained.tar) | 76.50% | 93.00% | 8.787 | 5.434 |
|[ResNet50_vc](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_vc_pretrained.tar) |78.35% | 94.03% | 9.013 | 5.463 |
|[ResNet50_vd](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_vd_pretrained.tar) | 79.12% | 94.44% | 9.058 | 5.510 |
|[ResNet50_vd_v2](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_vd_v2_pretrained.tar) | 79.84% | 94.93% | 9.058 | 5.510 |
|[ResNet101](http://paddle-imagenet-models-name.bj.bcebos.com/ResNet101_pretrained.tar) | 77.56% | 93.64% | 15.447 | 8.779 |
|[ResNet101_vd](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet101_vd_pretrained.tar) | 80.17% | 94.97% | 15.685 | 8.878 |
|[ResNet152](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet152_pretrained.tar) | 78.26% | 93.96% | 21.816 | 12.148 |
|[ResNet152_vd](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet152_vd_pretrained.tar) | 80.59% | 95.30% | 22.041 | 12.259 |
|[ResNet200_vd](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet200_vd_pretrained.tar) | 80.93% | 95.33% | 28.015 | 15.278 |
### ResNeXt Series
|model | top-1/top-5 accuracy(CV2) | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |
|[ResNeXt50_32x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt50_32x4d_pretrained.tar) | 77.75%/93.82% | 12.863 | 9.837 |
|[ResNeXt50_vd_32x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt50_vd_32x4d_pretrained.tar) | 79.56%/94.62% | 13.673 | 9.991 |
|[ResNeXt50_64x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt50_64x4d_pretrained.tar) | 78.43%/94.13% | 28.162 | 18.271 |
|[ResNeXt50_vd_64x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt50_vd_64x4d_pretrained.tar) | 80.12%/94.86% | 20.888 | 17.687 |
|[ResNeXt101_32x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt101_32x4d_pretrained.tar) | 78.65%/94.19% | 24.154 | 21.387 |
|[ResNeXt101_64x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt50_64x4d_pretrained.tar) | 78.43%/94.13% | 41.073 | 38.736 |
|[ResNeXt101_vd_64x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt101_vd_64x4d_pretrained.tar) | 80.78%/95.20% | 42.277 | 40.929 |
|[ResNeXt152_32x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt152_32x4d_pretrained.tar) | 78.98%/94.33% | 37.007 | 31.301 |
|[ResNeXt152_64x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt152_64x4d_pretrained.tar) | 79.51%/94.71% | 58.966 | 57.267 |
|Model | Top-1 | Top-5 | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |:-: |
|[ResNeXt50_32x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt50_32x4d_pretrained.tar) | 77.75% | 93.82% | 12.863 | 9.837 |
|[ResNeXt50_vd_32x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt50_vd_32x4d_pretrained.tar) | 79.56% | 94.62% | 13.673 | 9.991 |
|[ResNeXt50_64x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt50_64x4d_pretrained.tar) | 78.43% | 94.13% | 28.162 | 18.271 |
|[ResNeXt50_vd_64x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt50_vd_64x4d_pretrained.tar) | 80.12% | 94.86% | 20.888 | 17.687 |
|[ResNeXt101_32x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt101_32x4d_pretrained.tar) | 78.65% | 94.19% | 24.154 | 21.387 |
|[ResNeXt101_vd_32x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt101_vd_32x4d_pretrained.tar) | 80.33% | 95.12% | 24.701 | 18.032 |
|[ResNeXt101_64x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt50_64x4d_pretrained.tar) | 78.43% | 94.13% | 41.073 | 38.736 |
|[ResNeXt101_vd_64x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt101_vd_64x4d_pretrained.tar) | 80.78% | 95.20% | 42.277 | 40.929 |
|[ResNeXt152_32x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt152_32x4d_pretrained.tar) | 78.98% | 94.33% | 37.007 | 31.301 |
|[ResNeXt152_64x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt152_64x4d_pretrained.tar) | 79.51% | 94.71% | 58.966 | 57.267 |
|[ResNeXt152_vd_64x4d](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt152_vd_64x4d_pretrained.tar) | 81.08% | 95.34% | 60.947 | 49.117 |
### DenseNet Series
|model | top-1/top-5 accuracy(CV2) | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |
|[DenseNet121](https://paddle-imagenet-models-name.bj.bcebos.com/DenseNet121_pretrained.tar) | 75.66%/92.58% | 12.437 | 5.813 |
|[DenseNet161](https://paddle-imagenet-models-name.bj.bcebos.com/DenseNet161_pretrained.tar) | 78.57%/94.14% | 27.717 | 12.861 |
|[DenseNet169](https://paddle-imagenet-models-name.bj.bcebos.com/DenseNet169_pretrained.tar) | 76.81%/93.31% | 18.941 | 8.146 |
|[DenseNet201](https://paddle-imagenet-models-name.bj.bcebos.com/DenseNet201_pretrained.tar) | 77.63%/93.66% | 26.583 | 10.549 |
|[DenseNet264](https://paddle-imagenet-models-name.bj.bcebos.com/DenseNet264_pretrained.tar) | 77.96%/93.85% | 41.495 | 15.574 |
|Model | Top-1 | Top-5 | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |:-: |
|[DenseNet121](https://paddle-imagenet-models-name.bj.bcebos.com/DenseNet121_pretrained.tar) | 75.66% | 92.58% | 12.437 | 5.813 |
|[DenseNet161](https://paddle-imagenet-models-name.bj.bcebos.com/DenseNet161_pretrained.tar) | 78.57% | 94.14% | 27.717 | 12.861 |
|[DenseNet169](https://paddle-imagenet-models-name.bj.bcebos.com/DenseNet169_pretrained.tar) | 76.81% | 93.31% | 18.941 | 8.146 |
|[DenseNet201](https://paddle-imagenet-models-name.bj.bcebos.com/DenseNet201_pretrained.tar) | 77.63% | 93.66% | 26.583 | 10.549 |
|[DenseNet264](https://paddle-imagenet-models-name.bj.bcebos.com/DenseNet264_pretrained.tar) | 77.96% | 93.85% | 41.495 | 15.574 |
### DPN Series
|Model | Top-1 | Top-5 | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |:-: |
|[DPN68](https://paddle-imagenet-models-name.bj.bcebos.com/DPN68_pretrained.tar) | 76.78% | 93.43% | 18.446 | 6.324 |
|[DPN92](https://paddle-imagenet-models-name.bj.bcebos.com/DPN92_pretrained.tar) | 79.85% | 94.80% | 25.748 | 22.182 |
|[DPN98](https://paddle-imagenet-models-name.bj.bcebos.com/DPN98_pretrained.tar) | 80.59% | 95.10% | 29.421 | 13.657 |
|[DPN107](https://paddle-imagenet-models-name.bj.bcebos.com/DPN107_pretrained.tar) | 80.89% | 95.32% | 41.071 | 19.115 |
|[DPN131](https://paddle-imagenet-models-name.bj.bcebos.com/DPN131_pretrained.tar) | 80.70% | 95.14% | 41.179 | 18.278 |
### SENet Series
|model | top-1/top-5 accuracy(CV2) | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |
|[SE_ResNeXt50_32x4d](https://paddle-imagenet-models-name.bj.bcebos.com/SE_ResNeXt50_32x4d_pretrained.tar) | 78.44%/93.96% | 14.916 | 12.126 |
|[SE_ResNeXt101_32x4d](https://paddle-imagenet-models-name.bj.bcebos.com/SE_ResNeXt101_32x4d_pretrained.tar) | 79.12%/94.20% | 30.085 | 24.110 |
|[SE_154_vd](https://paddle-imagenet-models-name.bj.bcebos.com/SE_154_vd_pretrained.tar) | 81.40%/95.48% | 71.892 | 64.855 |
|Model | Top-1 | Top-5 | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |:-: |
|[SE_ResNet50_vd](https://paddle-imagenet-models-name.bj.bcebos.com/SE_ResNet50_vd_pretrained.tar) | 79.52% | 94.75% | 10.345 | 7.662 |
|[SE_ResNeXt50_32x4d](https://paddle-imagenet-models-name.bj.bcebos.com/SE_ResNeXt50_32x4d_pretrained.tar) | 78.44% | 93.96% | 14.916 | 12.126 |
|[SE_ResNeXt101_32x4d](https://paddle-imagenet-models-name.bj.bcebos.com/SE_ResNeXt101_32x4d_pretrained.tar) | 79.12% | 94.20% | 30.085 | 24.110 |
|[SENet_154_vd](https://paddle-imagenet-models-name.bj.bcebos.com/SENet_154_vd_pretrained.tar) | 81.40% | 95.48% | 71.892 | 64.855 |
### Inception Series
|model | top-1/top-5 accuracy(CV2) | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |
|[GoogLeNet](https://paddle-imagenet-models-name.bj.bcebos.com/GoogLeNet_pretrained.tar) | 70.70%/89.66% | 6.528 | 3.076 |
|[Xception_41](https://paddle-imagenet-models-name.bj.bcebos.com/Xception_41_pretrained.tar) | 79.30%/94.53% | 13.757 | 10.831 |
|[InceptionV4](https://paddle-imagenet-models-name.bj.bcebos.com/InceptionV4_pretrained.tar) | 80.77%/95.26% | 32.413 | 18.154 |
| Model | Top-1 | Top-5 | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |:-: |
|[GoogLeNet](https://paddle-imagenet-models-name.bj.bcebos.com/GoogLeNet_pretrained.tar) | 70.70% | 89.66% | 6.528 | 3.076 |
|[Xception41](https://paddle-imagenet-models-name.bj.bcebos.com/Xception41_pretrained.tar) | 79.30% | 94.53% | 13.757 | 10.831 |
|[Xception41_deeplab](https://paddle-imagenet-models-name.bj.bcebos.com/Xception41_deeplab_pretrained.tar) | 79.55% | 94.38% | 14.268 | 10.301 |
|[Xception65](https://paddle-imagenet-models-name.bj.bcebos.com/Xception65_pretrained.tar) | 81.00% | 95.49% | 19.216 | 15.981 |
|[Xception65_deeplab](https://paddle-imagenet-models-name.bj.bcebos.com/Xception65_deeplab_pretrained.tar) | 80.32% | 94.49% | 19.536 | 16.365 |
|[Xception71](https://paddle-imagenet-models-name.bj.bcebos.com/Xception71_pretrained.tar) | 81.11% | 95.45% | 23.291 | 18.974 |
|[InceptionV4](https://paddle-imagenet-models-name.bj.bcebos.com/InceptionV4_pretrained.tar) | 80.77% | 95.26% | 32.413 | 18.154 |
### DarkNet
|model | top-1/top-5 accuracy(CV2) | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |
|[DarkNet53](https://paddle-imagenet-models-name.bj.bcebos.com/DarkNet53_ImageNet1k_pretrained.tar) | 78.04%/94.05% | 11.969 | 7.153 |
### ResNeXt101_:wwsl Series
|model | top-1/top-5 accuracy(CV2) | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |
|[ResNeXt101_32x8d_wsl](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt101_32x8d_wsl_pretrained.tar) | 82.55%/96.74% | 33.310 | 27.648 |
|[ResNeXt101_32x16d_wsl](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt101_32x16d_wsl_pretrained.tar) | 84.24%/97.26% | 54.320 | 46.064 |
|[ResNeXt101_32x32d_wsl](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt101_32x32d_wsl_pretrained.tar) | 84.97%/97.59% | 97.734 | 87.961 |
|[ResNeXt101_32x48d_wsl](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt101_32x48d_wsl_pretrained.tar) | 85.37%/97.69% | 161.722 | |
|[Fix_ResNeXt101_32x48d_wsl](https://paddle-imagenet-models-name.bj.bcebos.com/Fix_ResNeXt101_32x48d_wsl_pretrained.tar) | 86.26%/97.97% | 236.091 | |
|Model | Top-1 | Top-5 | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |:-: |
|[DarkNet53](https://paddle-imagenet-models-name.bj.bcebos.com/DarkNet53_ImageNet1k_pretrained.tar) | 78.04% | 94.05% | 11.969 | 7.153 |
### ResNeXt101_wsl Series
|Model | Top-1 | Top-5 | Paddle Fluid inference time(ms) | Paddle TensorRT inference time(ms) |
|- |:-: |:-: |:-: |:-: |
|[ResNeXt101_32x8d_wsl](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt101_32x8d_wsl_pretrained.tar) | 82.55% | 96.74% | 33.310 | 27.648 |
|[ResNeXt101_32x16d_wsl](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt101_32x16d_wsl_pretrained.tar) | 84.24% | 97.26% | 54.320 | 46.064 |
|[ResNeXt101_32x32d_wsl](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt101_32x32d_wsl_pretrained.tar) | 84.97% | 97.59% | 97.734 | 87.961 |
|[ResNeXt101_32x48d_wsl](https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt101_32x48d_wsl_pretrained.tar) | 85.37% | 97.69% | 161.722 | |
|[Fix_ResNeXt101_32x48d_wsl](https://paddle-imagenet-models-name.bj.bcebos.com/Fix_ResNeXt101_32x48d_wsl_pretrained.tar) | 86.26% | 97.97% | 236.091 | |
## FAQ
......@@ -340,12 +363,14 @@ Enforce failed. Expected x_dims[1] == labels_dims[1], but received x_dims[1]:100
- ShuffleNetV2: [ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design](https://arxiv.org/abs/1807.11164), Ningning Ma, Xiangyu Zhang, Hai-Tao Zheng, Jian Sun
- MobileNetV1: [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://arxiv.org/abs/1704.04861), Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, Hartwig Adam
- MobileNetV2: [MobileNetV2: Inverted Residuals and Linear Bottlenecks](https://arxiv.org/pdf/1801.04381v4.pdf), Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, Liang-Chieh Chen
- MobileNetV3: [Searching for MobileNetV3](https://arxiv.org/pdf/1905.02244.pdf), Andrew Howard, Mark Sandler, Grace Chu, Liang-Chieh Chen, Bo Chen, Mingxing Tan, Weijun Wang, Yukun Zhu, Ruoming Pang, Vijay Vasudevan, Quoc V. Le, Hartwig Adam
- VGG: [Very Deep Convolutional Networks for Large-scale Image Recognition](https://arxiv.org/pdf/1409.1556), Karen Simonyan, Andrew Zisserman
- GoogLeNet: [Going Deeper with Convolutions](https://www.cs.unc.edu/~wliu/papers/GoogLeNet.pdf), Christian Szegedy1, Wei Liu2, Yangqing Jia
- Xception: [Xception: Deep Learning with Depthwise Separable Convolutions](https://arxiv.org/abs/1610.02357), Franc ̧ois Chollet
- InceptionV4: [Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning](https://arxiv.org/abs/1602.07261), Christian Szegedy, Sergey Ioffe, Vincent Vanhoucke, Alex Alemi
- DarkNet: [YOLOv3: An Incremental Improvement](https://pjreddie.com/media/files/papers/YOLOv3.pdf), Joseph Redmon, Ali Farhadi
- DenseNet: [Densely Connected Convolutional Networks](https://arxiv.org/abs/1608.06993), Gao Huang, Zhuang Liu, Laurens van der Maaten
- DPN: [Dual Path Networks](https://arxiv.org/pdf/1707.01629.pdf), Yunpeng Chen, Jianan Li, Huaxin Xiao, Xiaojie Jin, Shuicheng Yan, Jiashi Feng
- SqueezeNet: [SQUEEZENET: ALEXNET-LEVEL ACCURACY WITH 50X FEWER PARAMETERS AND <0.5MB MODEL SIZE](https://arxiv.org/abs/1602.07360), Forrest N. Iandola, Song Han, Matthew W. Moskewicz, Khalid Ashraf, William J. Dally, Kurt Keutzer
- ResNeXt101_wsl: [Exploring the Limits of Weakly Supervised Pretraining](https://arxiv.org/abs/1805.00932), Dhruv Mahajan, Ross Girshick, Vignesh Ramanathan, Kaiming He, Manohar Paluri, Yixuan Li, Ashwin Bharambe, Laurens van der Maaten
- Fix_ResNeXt101_wsl: [Fixing the train-test resolution discrepancy](https://arxiv.org/abs/1906.06423), Hugo Touvron, Andrea Vedaldi, Matthijs Douze, Herve ́ Je ́gou
......@@ -358,9 +383,10 @@ Enforce failed. Expected x_dims[1] == labels_dims[1], but received x_dims[1]:100
- 2019/04/01 **Stage3**: Update ResNet18, ResNet34, GoogLeNet, ShuffleNetV2
- 2019/06/12 **Stage4**:Update ResNet50_vc, ResNet50_vd, ResNet101_vd, ResNet152_vd, ResNet200_vd, SE154_vd InceptionV4, ResNeXt101_64x4d, ResNeXt101_vd_64x4d
- 2019/06/22 Update ResNet50_vd_v2
- 2019/07/02 **Stage5**: Update MobileNetV2_x0_5, ResNeXt50_32x4d, ResNeXt50_64x4d, Xception_41, ResNet101_vd
- 2019/07/02 **Stage5**: Update MobileNetV2_x0_5, ResNeXt50_32x4d, ResNeXt50_64x4d, Xception41, ResNet101_vd
- 2019/07/19 **Stage6**: Update ShuffleNetV2_x0_25, ShuffleNetV2_x0_33, ShuffleNetV2_x0_5, ShuffleNetV2_x1_0, ShuffleNetV2_x1_5, ShuffleNetV2_x2_0, MobileNetV2_x0_25, MobileNetV2_x1_5, MobileNetV2_x2_0, ResNeXt50_vd_64x4d, ResNeXt101_32x4d, ResNeXt152_32x4d
- 2019/08/01 **Stage7**: Update DarkNet53, DenseNet121. Densenet161, DenseNet169, DenseNet201, DenseNet264, SqueezeNet1_0, SqueezeNet1_1, ResNeXt50_vd_32x4d, ResNeXt152_64x4d, ResNeXt101_32x8d_wsl, ResNeXt101_32x16d_wsl, ResNeXt101_32x32d_wsl, ResNeXt101_32x48d_wsl, Fix_ResNeXt101_32x48d_wsl
- 2019/09/11 **Stage8**: Update ResNet18_vd,ResNet34_vd,MobileNetV1_x0_25,MobileNetV1_x0_5,MobileNetV1_x0_75,MobileNetV2_x0_75,MobilenNetV3_small_x1_0,DPN68,DPN92,DPN98,DPN107,DPN131,ResNeXt101_vd_32x4d,ResNeXt152_vd_64x4d,Xception65,Xception71,Xception41_deeplab,Xception65_deeplab,SE_ResNet50_vd
## Contribute
......
......@@ -13,23 +13,26 @@
#limitations under the License.
from .alexnet import AlexNet
from .mobilenet import MobileNet, MobileNetV1
from .mobilenet_v2 import MobileNetV2, MobileNetV2_x0_25, MobileNetV2_x0_5, MobileNetV2_x1_0, MobileNetV2_x1_5, MobileNetV2_x2_0, MobileNetV2_scale
from .mobilenet_v1 import MobileNetV1_x0_25, MobileNetV1_x0_5, MobileNetV1_x0_75, MobileNetV1
from .mobilenet_v2 import MobileNetV2_x0_25, MobileNetV2_x0_5, MobileNetV2_x0_75, MobileNetV2, MobileNetV2_x1_5, MobileNetV2_x2_0
from .mobilenet_v3 import MobileNetV3_small_x0_25, MobileNetV3_small_x0_5, MobileNetV3_small_x0_75, MobileNetV3_small_x1_0, MobileNetV3_small_x1_25, MobileNetV3_large_x0_25, MobileNetV3_large_x0_5, MobileNetV3_large_x0_75, MobileNetV3_large_x1_0, MobileNetV3_large_x1_25
from .googlenet import GoogLeNet
from .vgg import VGG11, VGG13, VGG16, VGG19
from .resnet import ResNet18, ResNet34, ResNet50, ResNet101, ResNet152
from .resnet_vc import ResNet50_vc, ResNet101_vc, ResNet152_vc
from .resnet_vd import ResNet50_vd, ResNet101_vd, ResNet152_vd, ResNet200_vd
from .resnet_vd import ResNet18_vd, ResNet34_vd, ResNet50_vd, ResNet101_vd, ResNet152_vd, ResNet200_vd
from .resnext import ResNeXt50_64x4d, ResNeXt101_64x4d, ResNeXt152_64x4d, ResNeXt50_32x4d, ResNeXt101_32x4d, ResNeXt152_32x4d
from .resnext_vd import ResNeXt50_vd_64x4d, ResNeXt101_vd_64x4d, ResNeXt152_vd_64x4d, ResNeXt50_vd_32x4d, ResNeXt101_vd_32x4d, ResNeXt152_vd_32x4d
from .inception_v4 import InceptionV4
from .se_resnet_vd import SE_ResNet18_vd, SE_ResNet34_vd, SE_ResNet50_vd, SE_ResNet101_vd, SE_ResNet152_vd, SE_ResNet200_vd
from .se_resnext import SE_ResNeXt50_32x4d, SE_ResNeXt101_32x4d, SE_ResNeXt152_32x4d
from .se_resnext_vd import SE_ResNeXt50_32x4d_vd, SE_ResNeXt101_32x4d_vd, SE_154_vd
from .se_resnext_vd import SE_ResNeXt50_vd_32x4d, SE_ResNeXt101_vd_32x4d, SENet154_vd
from .dpn import DPN68, DPN92, DPN98, DPN107, DPN131
from .shufflenet_v2_swish import ShuffleNetV2_swish, ShuffleNetV2_x0_5_swish, ShuffleNetV2_x1_0_swish, ShuffleNetV2_x1_5_swish, ShuffleNetV2_x2_0_swish
from .shufflenet_v2 import ShuffleNetV2_x0_25, ShuffleNetV2_x0_33, ShuffleNetV2_x0_5, ShuffleNetV2_x1_0, ShuffleNetV2_x1_5, ShuffleNetV2_x2_0, ShuffleNetV2
from .fast_imagenet import FastImageNet
from .xception import Xception_41, Xception_65, Xception_71
from .xception import Xception41, Xception65, Xception71
from .xception_deeplab import Xception41_deeplab , Xception65_deeplab , Xception71_deeplab
from .densenet import DenseNet121, DenseNet161, DenseNet169, DenseNet201, DenseNet264
from .squeezenet import SqueezeNet1_0, SqueezeNet1_1
from .darknet import DarkNet53
......
......@@ -37,7 +37,7 @@ class DPN(object):
args = self.get_net_args(self.layers)
bws = args['bw']
inc_sec = args['inc_sec']
rs = args['bw']
rs = args['r']
k_r = args['k_r']
k_sec = args['k_sec']
G = args['G']
......@@ -121,12 +121,10 @@ class DPN(object):
pool_type='avg', )
stdv = 0.01
param_attr = fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv))
fc6 = fluid.layers.fc(input=pool5,
size=class_dim,
param_attr=param_attr,
name="fc6")
param_attr=ParamAttr(initializer=fluid.initializer.Uniform(-stdv, stdv), name='fc_weights'),
bias_attr=ParamAttr(name='fc_offset'))
return fc6
......@@ -317,7 +315,7 @@ def DPN68():
def DPN92():
onvodel = DPN(layers=92)
model = DPN(layers=92)
return model
......
......@@ -20,14 +20,15 @@ import paddle.fluid as fluid
from paddle.fluid.initializer import MSRA
from paddle.fluid.param_attr import ParamAttr
__all__ = ['MobileNet', 'MobileNetV1']
__all__ = ['MobileNetV1', 'MobileNetV1_x0_25', 'MobileNetV1_x0_5', 'MobileNetV1_x0_75']
class MobileNet():
def __init__(self):
pass
class MobileNetV1():
def __init__(self, scale=1.0):
self.scale = scale
def net(self, input, class_dim=1000, scale=1.0):
def net(self, input, class_dim=1000):
scale = self.scale
# conv1: 112x112
input = self.conv_bn_layer(
input,
......@@ -194,6 +195,21 @@ class MobileNet():
return pointwise_conv
def MobileNetV1_x0_25():
model = MobileNetV1(scale=0.25)
return model
def MobileNetV1_x0_5():
model = MobileNetV1(scale=0.5)
return model
def MobileNetV1_x0_75():
model = MobileNetV1(scale=0.75)
return model
def MobileNetV1():
model = MobileNet()
model = MobileNetV1(scale=1.0)
return model
......@@ -20,21 +20,18 @@ from paddle.fluid.initializer import MSRA
from paddle.fluid.param_attr import ParamAttr
__all__ = [
'MobileNetV2', 'MobileNetV2_x0_25, '
'MobileNetV2_x0_5', 'MobileNetV2_x1_0', 'MobileNetV2_x1_5',
'MobileNetV2_x2_0', 'MobileNetV2_scale'
'MobileNetV2_x0_25', 'MobileNetV2_x0_5'
'MobileNetV2_x0_75', 'MobileNetV2', 'MobileNetV2_x1_5',
'MobileNetV2_x2_0',
]
class MobileNetV2():
def __init__(self, scale=1.0, change_depth=False):
def __init__(self, scale=1.0):
self.scale = scale
self.change_depth = change_depth
def net(self, input, class_dim=1000):
scale = self.scale
change_depth = self.change_depth
#if change_depth is True, the new depth is 1.4 times as deep as before.
bottleneck_params_list = [
(1, 16, 1, 1),
(6, 24, 2, 2),
......@@ -43,14 +40,6 @@ class MobileNetV2():
(6, 96, 3, 1),
(6, 160, 3, 2),
(6, 320, 1, 1),
] if change_depth == False else [
(1, 16, 1, 1),
(6, 24, 2, 2),
(6, 32, 5, 2),
(6, 64, 7, 2),
(6, 96, 5, 1),
(6, 160, 3, 2),
(6, 320, 1, 1),
]
#conv1
......@@ -221,11 +210,17 @@ def MobileNetV2_x0_5():
return model
def MobileNetV2_x1_0():
def MobileNetV2_x0_75():
model = MobileNetV2(scale=0.75)
return model
def MobileNetV2():
model = MobileNetV2(scale=1.0)
return model
def MobileNetV2_x1_5():
model = MobileNetV2(scale=1.5)
return model
......@@ -234,8 +229,3 @@ def MobileNetV2_x1_5():
def MobileNetV2_x2_0():
model = MobileNetV2(scale=2.0)
return model
def MobileNetV2_scale():
model = MobileNetV2(scale=1.2, change_depth=True)
return model
#copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
#
#Licensed under the Apache License, Version 2.0 (the "License");
#you may not use this file except in compliance with the License.
#You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
#Unless required by applicable law or agreed to in writing, software
#distributed under the License is distributed on an "AS IS" BASIS,
#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
#See the License for the specific language governing permissions and
#limitations under the License.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import paddle.fluid as fluid
from paddle.fluid.initializer import MSRA
from paddle.fluid.param_attr import ParamAttr
__all__ = ['MobileNetV3','MobileNetV3_small_x0_25', 'MobileNetV3_small_x0_5', 'MobileNetV3_small_x0_75',
'MobileNetV3_small_x1_0', 'MobileNetV3_small_x1_25', 'MobileNetV3_large_x0_25', 'MobileNetV3_large_x0_5',
'MobileNetV3_large_x0_75', 'MobileNetV3_large_x1_0', 'MobileNetV3_large_x1_25']
class MobileNetV3():
def __init__(self, scale=1.0, model_name='small'):
self.scale = scale
self.inplanes=16
if model_name == "large":
self.cfg = [
# k, exp, c, se, nl, s,
[3, 16, 16, False, 'relu', 1],
[3, 64, 24, False, 'relu', 2],
[3, 72, 24, False, 'relu', 1],
[5, 72, 40, True, 'relu', 2],
[5, 120, 40, True, 'relu', 1],
[5, 120, 40, True, 'relu', 1],
[3, 240, 80, False, 'hard_swish', 2],
[3, 200, 80, False, 'hard_swish', 1],
[3, 184, 80, False, 'hard_swish', 1],
[3, 184, 80, False, 'hard_swish', 1],
[3, 480, 112, True, 'hard_swish', 1],
[3, 672, 112, True, 'hard_swish', 1],
[5, 672, 160, True, 'hard_swish', 2],
[5, 960, 160, True, 'hard_swish', 1],
[5, 960, 160, True, 'hard_swish', 1],
]
self.cls_ch_squeeze = 960
self.cls_ch_expand = 1280
elif model_name == "small":
self.cfg = [
# k, exp, c, se, nl, s,
[3, 16, 16, True, 'relu', 2],
[3, 72, 24, False, 'relu', 2],
[3, 88, 24, False, 'relu', 1],
[5, 96, 40, True, 'hard_swish', 2],
[5, 240, 40, True, 'hard_swish', 1],
[5, 240, 40, True, 'hard_swish', 1],
[5, 120, 48, True, 'hard_swish', 1],
[5, 144, 48, True, 'hard_swish', 1],
[5, 288, 96, True, 'hard_swish', 2],
[5, 576, 96, True, 'hard_swish', 1],
[5, 576, 96, True, 'hard_swish', 1],
]
self.cls_ch_squeeze = 576
self.cls_ch_expand = 1280
else:
raise NotImplementedError
def net(self, input, class_dim=1000):
scale = self.scale
inplanes = self.inplanes
cfg = self.cfg
cls_ch_squeeze = self.cls_ch_squeeze
cls_ch_expand = self.cls_ch_expand
#conv1
conv = self.conv_bn_layer(input,
filter_size=3,
num_filters=int(scale*inplanes),
stride=2,
padding=1,
num_groups=1,
if_act=True,
act='hard_swish',
name='conv1')
i = 0
for layer_cfg in cfg:
conv = self.residual_unit(input=conv,
num_in_filter=inplanes,
num_mid_filter=int(scale*layer_cfg[1]),
num_out_filter=int(scale*layer_cfg[2]),
act=layer_cfg[4],
stride=layer_cfg[5],
filter_size=layer_cfg[0],
use_se=layer_cfg[3],
name='conv' + str(i+2))
inplanes = int(scale*layer_cfg[2])
i += 1
conv = self.conv_bn_layer(input=conv,
filter_size=1,
num_filters=int(scale*cls_ch_squeeze),
stride=1,
padding=0,
num_groups=1,
if_act=True,
act='hard_swish',
name='conv_last')
conv = fluid.layers.pool2d(input=conv, pool_type='avg', global_pooling=True, use_cudnn=False)
conv = fluid.layers.conv2d(
input=conv,
num_filters=cls_ch_expand,
filter_size=1,
stride=1,
padding=0,
act=None,
param_attr=ParamAttr(name='last_1x1_conv_weights'),
bias_attr=False)
conv = self.hard_swish(conv)
drop = fluid.layers.dropout(x=conv, dropout_prob=0.2)
out = fluid.layers.fc(input=drop,
size=class_dim,
param_attr=ParamAttr(name='fc_weights'),
bias_attr=ParamAttr(name='fc_offset'))
return out
def conv_bn_layer(self,
input,
filter_size,
num_filters,
stride,
padding,
num_groups=1,
if_act=True,
act=None,
name=None,
use_cudnn=True):
conv = fluid.layers.conv2d(
input=input,
num_filters=num_filters,
filter_size=filter_size,
stride=stride,
padding=padding,
groups=num_groups,
act=None,
use_cudnn=use_cudnn,
param_attr=ParamAttr(name=name+'_weights'),
bias_attr=False)
bn_name = name + '_bn'
bn = fluid.layers.batch_norm(input=conv,
param_attr = ParamAttr(name=bn_name+"_scale"),
bias_attr=ParamAttr(name=bn_name+"_offset"),
moving_mean_name=bn_name + '_mean',
moving_variance_name=bn_name + '_variance')
if if_act:
if act == 'relu':
bn = fluid.layers.relu(bn)
elif act == 'hard_swish':
bn = self.hard_swish(bn)
return bn
def hard_swish(self, x):
return x * fluid.layers.relu6(x+3) / 6.
def se_block(self, input, num_out_filter, ratio=4, name=None):
num_mid_filter = int(num_out_filter // ratio)
pool = fluid.layers.pool2d(
input=input, pool_type='avg', global_pooling=True, use_cudnn=False)
conv1 = fluid.layers.conv2d(input=pool, filter_size=1, num_filters=num_mid_filter, act='relu',
param_attr=ParamAttr(name=name+'_1_weights'),
bias_attr=ParamAttr(name=name+'_1_offset'))
conv2 = fluid.layers.conv2d(input=conv1, filter_size=1, num_filters=num_out_filter, act='hard_sigmoid',
param_attr=ParamAttr(name=name+'_2_weights'),
bias_attr=ParamAttr(name=name+'_2_offset'))
scale = fluid.layers.elementwise_mul(x=input, y=conv2, axis=0)
return scale
def residual_unit(self, input, num_in_filter, num_mid_filter, num_out_filter, stride,
filter_size, act=None, use_se=False, name=None):
first_conv = (num_out_filter != num_mid_filter)
input_data=input
if first_conv:
input = self.conv_bn_layer(input=input,
filter_size=1,
num_filters=num_mid_filter,
stride=1,
padding=0,
if_act=True,
act=act,
name=name + '_expand')
conv1 = self.conv_bn_layer(input=input,
filter_size=filter_size,
num_filters=num_mid_filter,
stride=stride,
padding=int((filter_size-1)//2),
if_act=True,
act=act,
num_groups=num_mid_filter,
use_cudnn=True,
name=name + '_depthwise')
if use_se:
conv1 = self.se_block(input=conv1, num_out_filter=num_mid_filter, name=name+'_se')
conv2 = self.conv_bn_layer(input=conv1,
filter_size=1,
num_filters=num_out_filter,
stride=1,
padding=0,
if_act=False,
name=name + '_linear')
if num_in_filter != num_out_filter or stride != 1:
return conv2
else:
return fluid.layers.elementwise_add(x=input_data, y=conv2, act=None)
def MobileNetV3_small_x0_25():
model=MobileNetV3(model_name='small', scale=0.25)
return model
def MobileNetV3_small_x0_5():
model=MobileNetV3(model_name='small', scale=0.5)
return model
def MobileNetV3_small_x0_75():
model=MobileNetV3(model_name='small', scale=0.75)
return model
def MobileNetV3_small_x1_0():
model=MobileNetV3(model_name='small', scale=1.0)
return model
def MobileNetV3_small_x1_25():
model=MobileNetV3(model_name='small', scale=1.25)
return model
def MobileNetV3_large_x0_25():
model=MobileNetV3(model_name='large', scale=0.25)
return model
def MobileNetV3_large_x0_5():
model=MobileNetV3(model_name='large', scale=0.5)
return model
def MobileNetV3_large_x0_75():
model=MobileNetV3(model_name='large', scale=0.75)
return model
def MobileNetV3_large_x1_0():
model=MobileNetV3(model_name='large', scale=1.0)
return model
def MobileNetV3_large_x1_25():
model=MobileNetV3(model_name='large', scale=1.25)
return model
#copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
#
#Licensed under the Apache License, Version 2.0 (the "License");
#you may not use this file except in compliance with the License.
#You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
#Unless required by applicable law or agreed to in writing, software
#distributed under the License is distributed on an "AS IS" BASIS,
#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
#See the License for the specific language governing permissions and
#limitations under the License.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import paddle
import paddle.fluid as fluid
import contextlib
bn_regularizer = fluid.regularizer.L2DecayRegularizer(regularization_coeff=0.0)
name_scope = ""
@contextlib.contextmanager
def scope(name):
global name_scope
bk = name_scope
name_scope = name_scope + name + '/'
yield
name_scope = bk
def max_pool(input, kernel, stride, padding):
data = fluid.layers.pool2d(input, pool_size=kernel, pool_type='max',
pool_stride=stride, pool_padding=padding)
return data
def group_norm(input, G, eps=1e-5, param_attr=None, bias_attr=None):
N, C, H, W = input.shape
if C % G != 0:
# print "group can not divide channle:", C, G
for d in range(10):
for t in [d, -d]:
if G + t <= 0: continue
if C % (G + t) == 0:
G = G + t
break
if C % G == 0:
# print "use group size:", G
break
assert C % G == 0
x = fluid.layers.group_norm(
input,
groups=G,
param_attr=param_attr,
bias_attr=bias_attr,
name=name_scope + 'group_norm')
return x
def bn(*args, **kargs):
with scope('BatchNorm'):
return fluid.layers.batch_norm(
*args,
epsilon=1e-3,
momentum=0.99,
param_attr=fluid.ParamAttr(
name=name_scope + 'gamma', regularizer=bn_regularizer),
bias_attr=fluid.ParamAttr(
name=name_scope + 'beta', regularizer=bn_regularizer),
moving_mean_name=name_scope + 'moving_mean',
moving_variance_name=name_scope + 'moving_variance',
**kargs)
def bn_relu(data):
return fluid.layers.relu(bn(data))
def relu(data):
return fluid.layers.relu(data)
def conv(*args, **kargs):
kargs['param_attr'] = name_scope + 'weights'
if 'bias_attr' in kargs and kargs['bias_attr']:
kargs['bias_attr'] = fluid.ParamAttr(
name=name_scope + 'biases',
regularizer=None,
initializer=fluid.initializer.ConstantInitializer(value=0.0))
else:
kargs['bias_attr'] = False
return fluid.layers.conv2d(*args, **kargs)
def deconv(*args, **kargs):
kargs['param_attr'] = name_scope + 'weights'
if 'bias_attr' in kargs and kargs['bias_attr']:
kargs['bias_attr'] = name_scope + 'biases'
else:
kargs['bias_attr'] = False
return fluid.layers.conv2d_transpose(*args, **kargs)
def seperate_conv(input, channel, stride, filter, dilation=1, act=None):
param_attr = fluid.ParamAttr(
name=name_scope + 'weights',
regularizer=fluid.regularizer.L2DecayRegularizer(
regularization_coeff=0.0),
initializer=fluid.initializer.TruncatedNormal(loc=0.0, scale=0.33))
with scope('depthwise'):
input = conv(
input,
input.shape[1],
filter,
stride,
groups=input.shape[1],
padding=(filter // 2) * dilation,
dilation=dilation,
use_cudnn=False,
param_attr=param_attr)
input = bn(input)
if act: input = act(input)
param_attr = fluid.ParamAttr(
name=name_scope + 'weights',
regularizer=None,
initializer=fluid.initializer.TruncatedNormal(loc=0.0, scale=0.06))
with scope('pointwise'):
input = conv(input, channel, 1, 1, groups=1, padding=0,
param_attr=param_attr)
input = bn(input)
if act: input = act(input)
return input
......@@ -23,7 +23,7 @@ import paddle.fluid as fluid
from paddle.fluid.param_attr import ParamAttr
__all__ = [
"ResNet", "ResNet50_vd", "ResNet101_vd", "ResNet152_vd", "ResNet200_vd"
"ResNet", "ResNet18_vd", "ResNet34_vd", "ResNet50_vd", "ResNet101_vd", "ResNet152_vd", "ResNet200_vd"
]
......@@ -35,11 +35,13 @@ class ResNet():
def net(self, input, class_dim=1000):
is_3x3 = self.is_3x3
layers = self.layers
supported_layers = [50, 101, 152, 200]
supported_layers = [18, 34, 50, 101, 152, 200]
assert layers in supported_layers, \
"supported layers are {} but input layer is {}".format(supported_layers, layers)
if layers == 50:
if layers == 18:
depth = [2, 2, 2, 2]
elif layers == 34 or layers == 50:
depth = [3, 4, 6, 3]
elif layers == 101:
depth = [3, 4, 23, 3]
......@@ -85,6 +87,7 @@ class ResNet():
pool_padding=1,
pool_type='max')
if layers >= 50:
for block in range(len(depth)):
for i in range(depth[block]):
if layers in [101, 152, 200] and block == 2:
......@@ -98,7 +101,17 @@ class ResNet():
input=conv,
num_filters=num_filters[block],
stride=2 if i == 0 and block != 0 else 1,
if_first=block == 0,
if_first=block==i==0,
name=conv_name)
else:
for block in range(len(depth)):
for i in range(depth[block]):
conv_name="res"+str(block+2)+chr(97+i)
conv = self.basic_block(
input=conv,
num_filters=num_filters[block],
stride=2 if i == 0 and block != 0 else 1,
if_first=block==i==0,
name=conv_name)
pool = fluid.layers.pool2d(
......@@ -156,7 +169,8 @@ class ResNet():
pool_size=2,
pool_stride=2,
pool_padding=0,
pool_type='avg')
pool_type='avg',
ceil_mode=True)
conv = fluid.layers.conv2d(
input=pool,
......@@ -186,11 +200,13 @@ class ResNet():
if if_first:
return self.conv_bn_layer(input, ch_out, 1, stride, name=name)
else:
return self.conv_bn_layer_new(
input, ch_out, 1, stride, name=name)
return self.conv_bn_layer_new(input, ch_out, 1, stride, name=name)
elif if_first:
return self.conv_bn_layer(input, ch_out, 1, stride, name=name)
else:
return input
def bottleneck_block(self, input, num_filters, stride, name, if_first):
conv0 = self.conv_bn_layer(
input=input,
......@@ -222,6 +238,38 @@ class ResNet():
return fluid.layers.elementwise_add(x=short, y=conv2, act='relu')
def basic_block(self, input, num_filters, stride, name, if_first):
conv0 = self.conv_bn_layer(
input=input,
num_filters=num_filters,
filter_size=3,
act='relu',
stride=stride,
name=name+"_branch2a")
conv1 = self.conv_bn_layer(
input=conv0,
num_filters=num_filters,
filter_size=3,
act=None,
name=name+"_branch2b")
short = self.shortcut(
input,
num_filters,
stride,
if_first=if_first,
name=name + "_branch1")
return fluid.layers.elementwise_add(x=short, y=conv1, act='relu')
def ResNet18_vd():
model=ResNet(layers=18, is_3x3=True)
return model
def ResNet34_vd():
model=ResNet(layers=34, is_3x3=True)
return model
def ResNet50_vd():
model = ResNet(layers=50, is_3x3=True)
return model
......
......@@ -158,7 +158,8 @@ class ResNeXt():
pool_size=2,
pool_stride=2,
pool_padding=0,
pool_type='avg')
pool_type='avg',
ceil_mode=True)
conv = fluid.layers.conv2d(
input=pool,
......
#copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
#
#Licensed under the Apache License, Version 2.0 (the "License");
#you may not use this file except in compliance with the License.
#You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
#Unless required by applicable law or agreed to in writing, software
#distributed under the License is distributed on an "AS IS" BASIS,
#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
#See the License for the specific language governing permissions and
#limitations under the License.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import paddle
import paddle.fluid as fluid
from paddle.fluid.param_attr import ParamAttr
import math
__all__ = ["SE_ResNet_vd", "SE_ResNet18_vd","SE_ResNet34_vd", "SE_ResNet50_vd", "SE_ResNet101_vd", "SE_ResNet152_vd",
"SE_ResNet200_vd"]
class SE_ResNet_vd():
def __init__(self, layers=50, is_3x3=False):
self.layers = layers
self.is_3x3 = is_3x3
def net(self, input, class_dim=1000):
is_3x3 = self.is_3x3
layers = self.layers
supported_layers = [18, 34, 50, 101, 152, 200]
assert layers in supported_layers, \
"supported layers are {} but input layer is {}".format(supported_layers, layers)
if layers == 18:
depth = [2, 2, 2, 2]
elif layers == 34 or layers == 50:
depth = [3, 4, 6, 3]
elif layers == 101:
depth = [3, 4, 23, 3]
elif layers == 152:
depth = [3, 8, 36, 3]
elif layers == 200:
depth = [3, 12, 48, 3]
num_filters = [64, 128, 256, 512]
reduction_ratio = 16
if is_3x3 == False:
conv = self.conv_bn_layer(
input=input, num_filters=64, filter_size=7, stride=2, act='relu')
else:
conv = self.conv_bn_layer(
input=input, num_filters=32, filter_size=3, stride=2, act='relu', name='conv1_1')
conv = self.conv_bn_layer(
input=conv, num_filters=32, filter_size=3, stride=1, act='relu', name='conv1_2')
conv = self.conv_bn_layer(
input=conv, num_filters=64, filter_size=3, stride=1, act='relu', name='conv1_3')
conv = fluid.layers.pool2d(
input=conv,
pool_size=3,
pool_stride=2,
pool_padding=1,
pool_type='max')
if layers >= 50:
for block in range(len(depth)):
for i in range(depth[block]):
if layers in [101, 152, 200] and block == 2:
if i == 0:
conv_name="res"+str(block+2)+"a"
else:
conv_name="res"+str(block+2)+"b"+str(i)
else:
conv_name="res"+str(block+2)+chr(97+i)
conv = self.bottleneck_block(
input=conv,
num_filters=num_filters[block],
stride=2 if i == 0 and block != 0 else 1,
if_first=block==i==0,
reduction_ratio=reduction_ratio,
name=conv_name)
else:
for block in range(len(depth)):
for i in range(depth[block]):
conv_name="res"+str(block+2)+chr(97+i)
conv = self.basic_block(
input=conv,
num_filters=num_filters[block],
stride=2 if i == 0 and block != 0 else 1,
if_first=block==i==0,
reduction_ratio=reduction_ratio,
name=conv_name)
pool = fluid.layers.pool2d(
input=conv, pool_size=7, pool_type='avg', global_pooling=True)
stdv = 1.0 / math.sqrt(pool.shape[1] * 1.0)
out = fluid.layers.fc(input=pool,
size=class_dim,
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv), name='fc6_weights'),
bias_attr=ParamAttr(name='fc6_offset'))
return out
def conv_bn_layer(self,
input,
num_filters,
filter_size,
stride=1,
groups=1,
act=None,
name=None):
conv = fluid.layers.conv2d(
input=input,
num_filters=num_filters,
filter_size=filter_size,
stride=stride,
padding=(filter_size - 1) // 2,
groups=groups,
act=None,
param_attr=ParamAttr(name=name + "_weights"),
bias_attr=False)
if name == "conv1":
bn_name = "bn_" + name
else:
bn_name = "bn" + name[3:]
return fluid.layers.batch_norm(input=conv,
act=act,
param_attr=ParamAttr(name=bn_name + '_scale'),
bias_attr=ParamAttr(bn_name + '_offset'),
moving_mean_name=bn_name + '_mean',
moving_variance_name=bn_name + '_variance')
def conv_bn_layer_new(self,
input,
num_filters,
filter_size,
stride=1,
groups=1,
act=None,
name=None):
pool = fluid.layers.pool2d(input=input,
pool_size=2,
pool_stride=2,
pool_padding=0,
pool_type='avg',
ceil_mode=True)
conv = fluid.layers.conv2d(
input=pool,
num_filters=num_filters,
filter_size=filter_size,
stride=1,
padding=(filter_size - 1) // 2,
groups=groups,
act=None,
param_attr=ParamAttr(name=name + "_weights"),
bias_attr=False)
if name == "conv1":
bn_name = "bn_" + name
else:
bn_name = "bn" + name[3:]
return fluid.layers.batch_norm(input=conv,
act=act,
param_attr=ParamAttr(name=bn_name + '_scale'),
bias_attr=ParamAttr(bn_name + '_offset'),
moving_mean_name=bn_name + '_mean',
moving_variance_name=bn_name + '_variance')
def shortcut(self, input, ch_out, stride, name, if_first=False):
ch_in = input.shape[1]
if ch_in != ch_out or stride != 1:
if if_first:
return self.conv_bn_layer(input, ch_out, 1, stride, name=name)
else:
return self.conv_bn_layer_new(input, ch_out, 1, stride, name=name)
elif if_first:
return self.conv_bn_layer(input, ch_out, 1, stride, name=name)
else:
return input
def bottleneck_block(self, input, num_filters, stride, name, if_first, reduction_ratio):
conv0 = self.conv_bn_layer(
input=input,
num_filters=num_filters,
filter_size=1,
act='relu',
name=name+"_branch2a")
conv1 = self.conv_bn_layer(
input=conv0,
num_filters=num_filters,
filter_size=3,
stride=stride,
act='relu',
name=name+"_branch2b")
conv2 =self.conv_bn_layer(
input=conv1,
num_filters=num_filters * 4,
filter_size=1,
act=None,
name=name+"_branch2c")
scale = self.squeeze_excitation(
input=conv2,
num_channels=num_filters * 4,
reduction_ratio=reduction_ratio,
name='fc_'+name)
short = self.shortcut(input, num_filters * 4, stride, if_first=if_first, name=name + "_branch1")
return fluid.layers.elementwise_add(x=short, y=scale, act='relu')
def basic_block(self, input, num_filters, stride, name, if_first, reduction_ratio):
conv0 = self.conv_bn_layer(input=input,
num_filters=num_filters,
filter_size=3,
act='relu',
stride=stride,
name=name+"_branch2a")
conv1 = self.conv_bn_layer(input=conv0,
num_filters=num_filters,
filter_size=3,
act=None,
name=name+"_branch2b")
scale = self.squeeze_excitation(
input=conv1,
num_channels=num_filters,
reduction_ratio=reduction_ratio,
name='fc_'+name)
short = self.shortcut(input,
num_filters,
stride,
if_first=if_first,
name=name + "_branch1")
return fluid.layers.elementwise_add(x=short, y=scale, act='relu')
def squeeze_excitation(self, input, num_channels, reduction_ratio, name=None):
pool = fluid.layers.pool2d(
input=input, pool_size=0, pool_type='avg', global_pooling=True)
stdv = 1.0 / math.sqrt(pool.shape[1] * 1.0)
squeeze = fluid.layers.fc(input=pool,
size=num_channels // reduction_ratio,
act='relu',
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(
-stdv, stdv),name=name+'_sqz_weights'),
bias_attr=ParamAttr(name=name+'_sqz_offset'))
stdv = 1.0 / math.sqrt(squeeze.shape[1] * 1.0)
excitation = fluid.layers.fc(input=squeeze,
size=num_channels,
act='sigmoid',
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=name+'_exc_weights'),
bias_attr=ParamAttr(name=name+'_exc_offset'))
scale = fluid.layers.elementwise_mul(x=input, y=excitation, axis=0)
return scale
def SE_ResNet18_vd():
model = SE_ResNet_vd(layers=18, is_3x3 = True)
return model
def SE_ResNet34_vd():
model = SE_ResNet_vd(layers=34, is_3x3 = True)
return model
def SE_ResNet50_vd():
model = SE_ResNet_vd(layers=50, is_3x3 = True)
return model
def SE_ResNet101_vd():
model = SE_ResNet_vd(layers=101, is_3x3 = True)
return model
def SE_ResNet152_vd():
model = SE_ResNet_vd(layers=152, is_3x3 = True)
return model
def SE_ResNet200_vd():
model = SE_ResNet_vd(layers=200, is_3x3 = True)
return model
......@@ -23,11 +23,11 @@ import paddle.fluid as fluid
from paddle.fluid.param_attr import ParamAttr
__all__ = [
"SE_ResNeXt", "SE_ResNeXt50_32x4d_vd", "SE_ResNeXt101_32x4d_vd", "SE154_vd"
"SE_ResNeXt_vd", "SE_ResNeXt50_32x4d_vd", "SE_ResNeXt101_32x4d_vd", "SENet154_vd"
]
class SE_ResNeXt():
class SE_ResNeXt_vd():
def __init__(self, layers=50):
self.layers = layers
......@@ -262,7 +262,8 @@ class SE_ResNeXt():
pool_size=2,
pool_stride=2,
pool_padding=0,
pool_type='avg')
pool_type='avg',
ceil_mode=True)
conv = fluid.layers.conv2d(
input=pool,
......@@ -312,16 +313,16 @@ class SE_ResNeXt():
return scale
def SE_ResNeXt50_32x4d_vd():
model = SE_ResNeXt(layers=50)
def SE_ResNeXt50_vd_32x4d():
model = SE_ResNeXt_vd(layers=50)
return model
def SE_ResNeXt101_32x4d_vd():
model = SE_ResNeXt(layers=101)
def SE_ResNeXt101_vd_32x4d():
model = SE_ResNeXt_vd(layers=101)
return model
def SE_154_vd():
model = SE_ResNeXt(layers=152)
def SENet154_vd():
model = SE_ResNeXt_vd(layers=152)
return model
......@@ -22,7 +22,7 @@ import math
import sys
from paddle.fluid.param_attr import ParamAttr
__all__ = ['Xception', 'Xception_41', 'Xception_65', 'Xception_71']
__all__ = ['Xception', 'Xception41', 'Xception65', 'Xception71']
class Xception(object):
......@@ -252,34 +252,29 @@ class Xception(object):
act=None,
param_attr=ParamAttr(name=name + "_weights"),
bias_attr=False,
use_cudnn=use_cudnn,
name=name + '.conv2d.output.1')
use_cudnn=use_cudnn)
bn_name = "bn_" + name
return fluid.layers.batch_norm(
input=conv,
act=act,
name=bn_name + '.output.1',
param_attr=ParamAttr(name=bn_name + '_scale'),
bias_attr=ParamAttr(bn_name + '_offset'),
moving_mean_name=bn_name + '_mean',
moving_variance_name=bn_name + '_variance')
def Xception_41():
"""Xception_41"""
def Xception41():
model = Xception(entry_flow_block_num=3, middle_flow_block_num=8)
return model
def Xception_65():
"""Xception_65"""
def Xception65():
model = Xception(entry_flow_block_num=3, middle_flow_block_num=16)
return model
def Xception_71():
"""Xception_71"""
def Xception71():
model = Xception(entry_flow_block_num=5, middle_flow_block_num=16)
return model
#copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
#
#Licensed under the Apache License, Version 2.0 (the "License");
#you may not use this file except in compliance with the License.
#You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
#Unless required by applicable law or agreed to in writing, software
#distributed under the License is distributed on an "AS IS" BASIS,
#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
#See the License for the specific language governing permissions and
#limitations under the License.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import contextlib
import paddle
import math
import paddle.fluid as fluid
from .model_libs import scope, name_scope
from .model_libs import bn, bn_relu, relu
from .model_libs import conv
from .model_libs import seperate_conv
__all__ = ['Xception41_deeplab', 'Xception65_deeplab', 'Xception71_deeplab']
def check_data(data, number):
if type(data) == int:
return [data] * number
assert len(data) == number
return data
def check_stride(s, os):
if s <= os:
return True
else:
return False
def check_points(count, points):
if points is None:
return False
else:
if isinstance(points, list):
return (True if count in points else False)
else:
return (True if count == points else False)
class Xception():
def __init__(self, backbone="xception_65"):
self.bottleneck_params = self.gen_bottleneck_params(backbone)
self.backbone = backbone
def gen_bottleneck_params(self, backbone='xception_65'):
if backbone == 'xception_65':
bottleneck_params = {
"entry_flow": (3, [2, 2, 2], [128, 256, 728]),
"middle_flow": (16, 1, 728),
"exit_flow": (2, [2, 1],[[728, 1024, 1024], [1536, 1536, 2048]])
}
elif backbone == 'xception_41':
bottleneck_params = {
"entry_flow": (3, [2, 2, 2], [128, 256, 728]),
"middle_flow": (8, 1, 728),
"exit_flow": (2, [2, 1],[[728, 1024, 1024], [1536, 1536, 2048]])
}
elif backbone == 'xception_71':
bottleneck_params = {
"entry_flow": (5, [2, 1, 2, 1, 2], [128, 256, 256, 728, 728]),
"middle_flow": (16, 1, 728),
"exit_flow": (2, [2, 1],[[728, 1024, 1024], [1536, 1536, 2048]])
}
else:
raise Exception("xception backbont only support xception_41/xception_65/xception_71")
return bottleneck_params
def net(self,
input,
output_stride=32,
class_dim=1000,
end_points=None,
decode_points=None):
self.stride = 2
self.block_point = 0
self.output_stride = output_stride
self.decode_points = decode_points
self.short_cuts = dict()
with scope(self.backbone):
# Entry flow
data = self.entry_flow(input)
if check_points(self.block_point, end_points):
return data, self.short_cuts
# Middle flow
data = self.middle_flow(data)
if check_points(self.block_point, end_points):
return data, self.short_cuts
# Exit flow
data = self.exit_flow(data)
if check_points(self.block_point, end_points):
return data, self.short_cuts
data = fluid.layers.reduce_mean(data, [2, 3], keep_dim=True)
data = fluid.layers.dropout(data, 0.5)
stdv = 1.0 / math.sqrt(data.shape[1] * 1.0)
with scope("logit"):
out = fluid.layers.fc(input=data, size=class_dim,
param_attr=fluid.param_attr.ParamAttr(name='fc_weights',
initializer=fluid.initializer.Uniform(-stdv, stdv)),
bias_attr=fluid.param_attr.ParamAttr(name='fc_bias'))
return out
def entry_flow(self, data):
param_attr = fluid.ParamAttr(
name=name_scope + 'weights',
regularizer=None,
initializer=fluid.initializer.TruncatedNormal(loc=0.0, scale=0.09))
with scope("entry_flow"):
with scope("conv1"):
data = bn_relu(conv(data, 32, 3, stride=2, padding=1, param_attr=param_attr))
with scope("conv2"):
data = bn_relu(conv(data, 64, 3, stride=1, padding=1, param_attr=param_attr))
# get entry flow params
block_num = self.bottleneck_params["entry_flow"][0]
strides = self.bottleneck_params["entry_flow"][1]
chns = self.bottleneck_params["entry_flow"][2]
strides = check_data(strides, block_num)
chns = check_data(chns, block_num)
# params to control your flow
s = self.stride
block_point = self.block_point
output_stride = self.output_stride
with scope("entry_flow"):
for i in range(block_num):
block_point = block_point + 1
with scope("block" + str(i + 1)):
stride = strides[i] if check_stride(s*strides[i], output_stride) else 1
data, short_cuts = self.xception_block(data, chns[i], [1, 1, stride])
s = s * stride
if check_points(block_point, self.decode_points):
self.short_cuts[block_point] = short_cuts[1]
self.stride = s
self.block_point = block_point
return data
def middle_flow(self, data):
block_num = self.bottleneck_params["middle_flow"][0]
strides = self.bottleneck_params["middle_flow"][1]
chns = self.bottleneck_params["middle_flow"][2]
strides = check_data(strides, block_num)
chns = check_data(chns, block_num)
# params to control your flow
s = self.stride
block_point = self.block_point
output_stride = self.output_stride
with scope("middle_flow"):
for i in range(block_num):
block_point = block_point + 1
with scope("block" + str(i + 1)):
stride = strides[i] if check_stride(s*strides[i], output_stride) else 1
data, short_cuts = self.xception_block(data, chns[i], [1, 1, strides[i]], skip_conv=False)
s = s * stride
if check_points(block_point, self.decode_points):
self.short_cuts[block_point] = short_cuts[1]
self.stride = s
self.block_point = block_point
return data
def exit_flow(self, data):
block_num = self.bottleneck_params["exit_flow"][0]
strides = self.bottleneck_params["exit_flow"][1]
chns = self.bottleneck_params["exit_flow"][2]
strides = check_data(strides, block_num)
chns = check_data(chns, block_num)
assert(block_num==2)
# params to control your flow
s = self.stride
block_point = self.block_point
output_stride = self.output_stride
with scope("exit_flow"):
with scope('block1'):
block_point += 1
stride = strides[0] if check_stride(s*strides[0], output_stride) else 1
data, short_cuts = self.xception_block(data, chns[0], [1, 1, stride])
s = s * stride
if check_points(block_point, self.decode_points):
self.short_cuts[block_point] = short_cuts[1]
with scope('block2'):
block_point += 1
stride = strides[1] if check_stride(s*strides[1], output_stride) else 1
data, short_cuts = self.xception_block(
data, chns[1], [1, 1, stride],
dilation=2,
has_skip=False,
activation_fn_in_separable_conv=True)
s = s * stride
if check_points(block_point, self.decode_points):
self.short_cuts[block_point] = short_cuts[1]
self.stride = s
self.block_point = block_point
return data
def xception_block(self,
input,
channels,
strides=1,
filters=3,
dilation=1,
skip_conv=True,
has_skip=True,
activation_fn_in_separable_conv=False):
repeat_number = 3
channels = check_data(channels, repeat_number)
filters = check_data(filters, repeat_number)
strides = check_data(strides, repeat_number)
data = input
results = []
for i in range(repeat_number):
with scope('separable_conv' + str(i + 1)):
if not activation_fn_in_separable_conv:
data = relu(data)
data = seperate_conv(
data,
channels[i],
strides[i],
filters[i],
dilation=dilation)
else:
data = seperate_conv(
data,
channels[i],
strides[i],
filters[i],
dilation=dilation,
act=relu)
results.append(data)
if not has_skip:
return data, results
if skip_conv:
param_attr = fluid.ParamAttr(
name=name_scope + 'weights',
regularizer=None,
initializer=fluid.initializer.TruncatedNormal(loc=0.0, scale=0.09))
with scope('shortcut'):
skip = bn(
conv(input, channels[-1], 1, strides[-1], groups=1,
padding=0, param_attr=param_attr))
else:
skip = input
return data + skip, results
def Xception41_deeplab():
model = Xception("xception_41")
return model
def Xception65_deeplab():
model = Xception("xception_65")
return model
def Xception71_deeplab():
model = Xception("xception_71")
return model
#DPN107
python train.py \
--model=DPN107 \
--batch_size=256 \
--total_images=1281167 \
--image_shape=3,224,224 \
--class_dim=1000 \
--lr_strategy=cosine_decay \
--lr=0.1 \
--num_epochs=200 \
--model_save_dir=output/ \
--l2_decay=1e-4 \
--use_mixup=True \
--use_label_smoothing=True \
--label_smoothing_epsilon=0.1
#DPN131
python train.py \
--model=DPN131 \
--batch_size=256 \
--total_images=1281167 \
--image_shape=3,224,224 \
--class_dim=1000 \
--lr_strategy=cosine_decay \
--lr=0.1 \
--num_epochs=200 \
--model_save_dir=output/ \
--l2_decay=1e-4 \
--use_mixup=True \
--use_label_smoothing=True \
--label_smoothing_epsilon=0.1
#DPN68
python train.py \
--model=DPN68 \
--batch_size=256 \
--total_images=1281167 \
--image_shape=3,224,224 \
--class_dim=1000 \
--lr_strategy=cosine_decay \
--lr=0.1 \
--num_epochs=200 \
--model_save_dir=output/ \
--l2_decay=1e-4 \
--use_mixup=True \
--use_label_smoothing=True \
--label_smoothing_epsilon=0.1
#DPN92
python train.py \
--model=DPN92 \
--batch_size=256 \
--total_images=1281167 \
--image_shape=3,224,224 \
--class_dim=1000 \
--lr_strategy=cosine_decay \
--lr=0.1 \
--num_epochs=200 \
--model_save_dir=output/ \
--l2_decay=1e-4 \
--use_mixup=True \
--use_label_smoothing=True \
--label_smoothing_epsilon=0.1
#DPN98
python train.py \
--model=DPN98 \
--batch_size=256 \
--total_images=1281167 \
--image_shape=3,224,224 \
--class_dim=1000 \
--lr_strategy=cosine_decay \
--lr=0.1 \
--num_epochs=200 \
--model_save_dir=output/ \
--l2_decay=1e-4 \
--use_mixup=True \
--use_label_smoothing=True \
--label_smoothing_epsilon=0.1
......@@ -7,7 +7,7 @@ export FLAGS_fraction_of_gpu_memory_to_use=0.98
python train.py \
--model=MobileNetV1 \
--model=MobileNetV1_x1_0 \
--batch_size=256 \
--total_images=1281167 \
--class_dim=1000 \
......
export CUDA_VISIBLE_DEVICES=0,1,2,3
export FLAGS_fast_eager_deletion_mode=1
export FLAGS_eager_delete_tensor_gb=0.0
export FLAGS_fraction_of_gpu_memory_to_use=0.98
python train.py \
--model=MobileNetV1_x0_25 \
--batch_size=256 \
--total_images=1281167 \
--class_dim=1000 \
--image_shape=3,224,224 \
--model_save_dir=output/ \
--lr_strategy=piecewise_decay \
--num_epochs=120 \
--lr=0.1 \
--l2_decay=3e-5 \
export CUDA_VISIBLE_DEVICES=0,1,2,3
export FLAGS_fast_eager_deletion_mode=1
export FLAGS_eager_delete_tensor_gb=0.0
export FLAGS_fraction_of_gpu_memory_to_use=0.98
python train.py \
--model=MobileNetV1_x0_5 \
--batch_size=256 \
--total_images=1281167 \
--class_dim=1000 \
--image_shape=3,224,224 \
--model_save_dir=output/ \
--lr_strategy=piecewise_decay \
--num_epochs=120 \
--lr=0.1 \
--l2_decay=3e-5 \
export CUDA_VISIBLE_DEVICES=0,1,2,3
export FLAGS_fast_eager_deletion_mode=1
export FLAGS_eager_delete_tensor_gb=0.0
export FLAGS_fraction_of_gpu_memory_to_use=0.98
python train.py \
--model=MobileNetV1_x0_75 \
--batch_size=256 \
--total_images=1281167 \
--class_dim=1000 \
--image_shape=3,224,224 \
--model_save_dir=output/ \
--lr_strategy=piecewise_decay \
--num_epochs=120 \
--lr=0.1 \
--l2_decay=3e-5 \
......@@ -7,7 +7,7 @@ export FLAGS_fraction_of_gpu_memory_to_use=0.98
python train.py \
--model=MobileNetV2 \
--model=MobileNetV2_x1_0 \
--batch_size=500 \
--total_images=1281167 \
--class_dim=1000 \
......
export CUDA_VISIBLE_DEVICES=0,1,2,3
export FLAGS_fast_eager_deletion_mode=1
export FLAGS_eager_delete_tensor_gb=0.0
export FLAGS_fraction_of_gpu_memory_to_use=0.98
python train.py \
--model=MobileNetV2_x0_75 \
--batch_size=256 \
--total_images=1281167 \
--class_dim=1000 \
--image_shape=3,224,224 \
--model_save_dir=output/ \
--lr_strategy=cosine_decay \
--num_epochs=240 \
--lr=0.045 \
--l2_decay=4e-5
#ResNeXt101_32x4d
#Training details
#Missed
export CUDA_VISIBLE_DEVICES=0,1,2,3
export FLAGS_fast_eager_deletion_mode=1
export FLAGS_eager_delete_tensor_gb=0.0
export FLAGS_fraction_of_gpu_memory_to_use=0.98
#ResNeXt101_32x4d
python train.py \
--model=ResNeXt101_32x4d \
--batch_size=256 \
......
#Training details
#Missed
export CUDA_VISIBLE_DEVICES=0,1,2,3
export FLAGS_fast_eager_deletion_mode=1
export FLAGS_eager_delete_tensor_gb=0.0
export FLAGS_fraction_of_gpu_memory_to_use=0.98
#ResNeXt101_vd_32x4d
python train.py \
--model=ResNeXt101_vd_32x4d \
--batch_size=256 \
--total_images=1281167 \
--image_shape=3,224,224 \
--class_dim=1000 \
--lr_strategy=cosine_decay \
--lr=0.1 \
--num_epochs=200 \
--model_save_dir=output/ \
--l2_decay=1e-4 \
--use_mixup=True \
--use_label_smoothing=True \
--label_smoothing_epsilon=0.1
#ResNeXt152_vd_64x4d
python train.py \
--model=ResNeXt152_vd_64x4d \
--batch_size=256 \
--total_images=1281167 \
--image_shape=3,224,224 \
--class_dim=1000 \
--lr_strategy=cosine_decay \
--lr=0.1 \
--num_epochs=200 \
--model_save_dir=output/ \
--l2_decay=1e-4 \
--use_mixup=True \
--use_label_smoothing=True \
--label_smoothing_epsilon=0.1 \
export CUDA_VISIBLE_DEVICES=0,1,2,3
export FLAGS_fast_eager_deletion_mode=1
export FLAGS_eager_delete_tensor_gb=0.0
export FLAGS_fraction_of_gpu_memory_to_use=0.98
python train.py \
--model=ResNet18_vd \
--batch_size=256 \
--total_images=1281167 \
--image_shape=3,224,224 \
--class_dim=1000 \
--lr_strategy=cosine_decay \
--lr=0.1 \
--num_epochs=200 \
--model_save_dir=output/ \
--l2_decay=7e-5 \
--use_mixup=True \
--use_label_smoothing=True \
--label_smoothing_epsilon=0.1
export CUDA_VISIBLE_DEVICES=0,1,2,3
export FLAGS_fast_eager_deletion_mode=1
export FLAGS_eager_delete_tensor_gb=0.0
export FLAGS_fraction_of_gpu_memory_to_use=0.98
python train.py \
--model=ResNet34_vd \
--batch_size=256 \
--total_images=1281167 \
--image_shape=3,224,224 \
--class_dim=1000 \
--lr_strategy=cosine_decay \
--lr=0.1 \
--num_epochs=200 \
--model_save_dir=output/ \
--l2_decay=7e-5 \
--use_mixup=True \
--use_label_smoothing=True \
--label_smoothing_epsilon=0.1
......@@ -7,7 +7,7 @@ export FLAGS_fraction_of_gpu_memory_to_use=0.98
#SE_154
python train.py \
--model=SE_154_vd \
--model=SENet154_vd \
--batch_size=256 \
--total_images=1281167 \
--image_shape=3,224,224 \
......
#SE_ResNet50_vd
python train.py \
--model=SE_ResNet50_vd \
--batch_size=256 \
--total_images=1281167 \
--image_shape=3,224,224 \
--class_dim=1000 \
--lr_strategy=cosine_decay \
--lr=0.1 \
--num_epochs=200 \
--model_save_dir=output/ \
--l2_decay=1e-4 \
--use_mixup=True \
--use_label_smoothing=True \
--label_smoothing_epsilon=0.1 \
python train.py \
--model=Xception_41 \
--model=Xception41 \
--batch_size=256 \
--total_images=1281167 \
--image_shape=3,224,224 \
--image_shape=3,299,299 \
--class_dim=1000 \
--lr_strategy=cosine_decay \
--lr=0.045 \
......
#Xception41_deeplab
python train.py \
--model=Xception41_deeplab \
--batch_size=256 \
--total_images=1281167 \
--image_shape=3,299,299 \
--class_dim=1000 \
--lr_strategy=cosine_decay \
--lr=0.045 \
--num_epochs=120 \
--model_save_dir=output/ \
--l2_decay=1e-4 \
--resize_short_size=320
#Xception65
python train.py \
--model=Xception65 \
--batch_size=256 \
--total_images=1281167 \
--image_shape=3,299,299 \
--class_dim=1000 \
--lr_strategy=cosine_decay \
--lr=0.1 \
--num_epochs=200 \
--model_save_dir=output/ \
--l2_decay=1e-4 \
--use_mixup=True \
--resize_short_size=320 \
--use_label_smoothing=True \
--label_smoothing_epsilon=0.1 \
#Xception65_deeplab
python train.py \
--model=Xception65_deeplab \
--batch_size=256 \
--total_images=1281167 \
--image_shape=3,299,299 \
--class_dim=1000 \
--lr_strategy=cosine_decay \
--lr=0.045 \
--num_epochs=120 \
--model_save_dir=output/ \
--l2_decay=1e-4 \
--resize_short_size=320
#Xception71
python train.py \
--model=Xception71 \
--batch_size=256 \
--total_images=1281167 \
--image_shape=3,299,299 \
--class_dim=1000 \
--lr_strategy=cosine_decay \
--lr=0.1 \
--num_epochs=200 \
--model_save_dir=output/ \
--l2_decay=1e-4 \
--use_mixup=True \
--resize_short_size=320 \
--use_label_smoothing=True \
--label_smoothing_epsilon=0.1 \
......@@ -123,7 +123,7 @@ def parse_args():
#add_arg('use_fp16', bool, False, "Whether to enable half precision training with fp16." )
#add_arg('scale_loss', float, 1.0, "The value of scale_loss for fp16." )
add_arg('use_label_smoothing', bool, False, "Whether to use label_smoothing")
add_arg('label_smoothing_epsilon', float, 0.2, "The value of label_smoothing_epsilon parameter")
add_arg('label_smoothing_epsilon', float, 0.1, "The value of label_smoothing_epsilon parameter")
#NOTE: (2019/08/08) temporary disable use_distill
#add_arg('use_distill', bool, False, "Whether to use distill")
add_arg('random_seed', int, None, "random seed")
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册