2020.11.4
- Support model conversion from dynamic computational graph of PaddlePaddle to ONNX.
- Reconfigure the code structure to support different versions of Paddlepaddle.
- Improve the coverage rate of ONNX Opset, and Opset 9,10 and 11 will be stably supported.
2020.9.21
- Support export to ONNX Opset 9,10 and 11.
- Support new convertable ops: swish, floor, uniform_random, abs, instance_norm, clip, tanh, log, norm and pad2d.
2019.09.25
- Add newly supported models: SE_ResNet50_vd,SqueezeNet1_0,SE_ResNext50_32x4d,Xception41,VGG16,InceptionV4 and YoloV3.
- Solve the incompatibility of Paddle2ONNX v0.1 with ONNX.
2019.08.20
- Solve the incompatibility of preview version with PaddlePaddle and ONNX.
- Support mainstream models of image classification and object detection models.
- Unify the interfaces and support PIP installation.