FAQ
Python library
Is the PyEDDL develop by the same team?
Sort of. We are all part of the DeepHealth project and, although each team is specialized in developing and supporting different libraries, we all contribute to the other projects so that the whole DeepHealth ecosystem can work smoothly.
Contributions
Can I contribute?
Yes, but first you need to open a new issue to explain and discuss your contribution.
Installation & Support
Do you have X feature supported?
We have two markdown where you can check the layers and operations that we currently support:
Do you have a Dockerfile prepared?
Yes, it’s on the root directory of the repo.
# Build the image
cd eddl/
docker build -f Dockerfile .
# Enter in the terminal
docker exec -it [container-id] bash
Performance
Is it faster than PyTorch/TensorFlow/etc?
Check our benchmakrs: EDDL benchmarks
Summary:
50% faster than PyTorch
Similar performance as Keras
Is it more memory-efficient than PyTorch/TensorFlow/etc?
Depends on your memory setting. You can take a look at our benchmarks: EDDL benchmarks
Problems with ONNX models
I cannot import a *.onnx model
If a *.onnx model cannot be imported by the EDDL, you can try to simplifying it. (More on: my document)
My *.onnx model is not exported correctly
Exporting an *.onnx is hard since each framework has its way of importing/exporting *.onnx files. Due to this problematic, we try to make sure that the EDDL can import/export with itself, ONNX Runtime, and then, we focus our efforts on the other frameworks.