European Distributed Deep Learning (EDDL) library. A general-purpose library initially developed to cover deep learning needs in healthcare use cases within the DeepHealth project.
Many ONNX models have dynamic input shapes such as 3xHxW that we cannot load.
Due to our static computational graph we cannot support these dynamic shapes, so I propose that ONNX should ask for the input shape if this is indeed required (dynamic input). Else, nothing.
Examples:
// Not dynamic input shape => Okay
Net *net = import_net_from_onnx_file(model_path);
// Dynamic input shape + No given input shape => Error message asking for an input shape
Net *net = import_net_from_onnx_file(model_path);
// Dynamic input shape + Given input shape => Okay
Net *net = import_net_from_onnx_file(model_path, {in_channels, in_height, in_width});
Many ONNX models have dynamic input shapes such as
3xHxW
that we cannot load.Due to our static computational graph we cannot support these dynamic shapes, so I propose that ONNX should ask for the input shape if this is indeed required (dynamic input). Else, nothing.
Examples: