FAQ

Python library

Is there a Python version?

Yes, the PyEDDL is the EDDL version for the Python lovers.

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:

I need X feature, can you add it?

Maybe… Open a new issue and we will happily discuss it.

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.