Paddle2ONNX supports converting PaddlePaddle model to ONNX format.
Due to the differences between frameworks, some models may not be supported. If you meet any problem such as converting failure or inference error，you can raise a issue in ISSUE.
Comprehensive coverage of image classification models，now we support the whole series model in PaddlClas release/2.1.
| Models | Source |
|---|---|
| ResNet series | PaddleClas |
| Mobile series | PaddleClas |
| SEResNeXt and Res2Net series | PaddleClas |
| DPN and DenseNet series | PaddleClas |
| HRNet series | PaddleClas |
| Inception series | PaddleClas |
| EfficientNet and ResNeXt101_wsl series | PaddleClas |
| ResNeSt and RegNet series | PaddleClas |
Support CRNN(Text Detection Model), DB(Text Recognition Model) and Text Angle Classification Model. Test models are form PaddleOCR dygraph branch.
| Models | Source |
|---|---|
| Chinese and English ultra-lightweight OCR model (9.4M) | PaddleOCR |
| Chinese and English general OCR model (143.4M) | PaddleOCR |
Support UNet, HRNet, DeepLab and so on. Test models are from PaddleSeg release/v2.0.0-rc branch。
| Models | Source |
|---|---|
| BiSeNet | PaddleSeg |
| DANet | PaddleSeg |
| DeepLabv3 | PaddleSeg |
| Deeplabv3P | PaddleSeg |
| FCN | PaddleSeg |
| GCNet | PaddleSeg |
| OCRNet | PaddleSeg |
| UNet | PaddleSeg |
Support 8 object detection archtectures. Test models are from PaddleDetection release/2.1