该脚本转换工具可将PyTorch脚本转换成MindSpore代码。
序号 |
模型 |
原始训练工程代码链接 |
备注 |
|
---|---|---|---|---|
1 |
ResNet系列 |
ResNet18 |
https://github.com/pytorch/examples/tree/master/imagenet |
|
2 |
ResNet34 |
|||
3 |
ResNet50 |
|||
4 |
ResNet101 |
|||
5 |
ResNet152 |
|||
6 |
ResNeXt-50(32x4d) |
|||
7 |
ResNeXt-101(32x8d) |
|||
8 |
Wide ResNet-50-2 |
|||
9 |
Wide ResNet-101-2 |
|||
10 |
ShuffleNet系列 |
ShuffleNetV2 |
https://github.com/megvii-model/ShuffleNet-Series |
- |
11 |
ShuffleNetV2+ |
|||
12 |
BERT |
https://github.com/codertimo/BERT-pytorch |
|
|
13 |
UNet |
https://github.com/milesial/Pytorch-Unet |
MindSpore暂不支持ReduceLROnPlateau,需要替换成其他支持的scheduler。 |
|
14 |
ViT |
https://github.com/jeonsworld/ViT-pytorch |
|
|
15 |
Res2Net |
https://github.com/Res2Net/Res2Net-ImageNet-Training |
暂不支持torchvision.models相关接口的转换,需做以下操作。 修改原始工程:
|
|
16 |
EfficientNet系列 |
EfficientNet-B0 |
https://github.com/lukemelas/EfficientNet-PyTorch |
- |
17 |
EfficientNet-B1 |
|||
18 |
EfficientNet-B2 |
|||
19 |
EfficientNet-B3 |
|||
20 |
EfficientNet-B4 |
|||
21 |
EfficientNet-B5 |
|||
22 |
EfficientNet-B6 |
|||
23 |
EfficientNet-B7 |
|||
24 |
EfficientNet-B8 |
|||
25 |
pytorch-cifar系列 |
VGG19 |
https://github.com/kuangliu/pytorch-cifar |
|
26 |
PreActResNet18 |
|||
27 |
DenseNet121 |
|||
28 |
ResNeXt29_2x64d |
|||
29 |
MobileNet |
|||
30 |
MobileNetV2 |
|||
31 |
SENet18 |
|||
32 |
ShuffleNetG2 |
|||
33 |
GoogleNet |
|||
34 |
DPN92 |
pip3 install pandas #pandas版本号需大于或等于1.2.4 pip3 install libcst