A repository for storing models that have been inter-converted between various frameworks. Supported frameworks are TensorFlow, PyTorch, ONNX, OpenVINO, TFJS, TFTRT, TensorFlowLite (Float32/16/INT8), EdgeTPU, CoreML.
As already discussed in #389 this PR adds the following improvements to the single pose example:
removed yolox example code
added strides parameter to avoid onnx dependency if not needed
added DirectML and CoreML execution providers for onnxruntime
added openvino support as runtime alternative (and a --runtime argument)
added upscaling of keypoints and masks to the size of the input image (if it does not match the input size)
removed duplicate indices from resulting keypoints list
Maybe it makes sense to extract the methods and libraries into a util module to not have duplicate code in the single- and multi-pose example. For now, this PR already helps to run the models on the CPU (and MacOS / Windows) as well.
As already discussed in #389 this PR adds the following improvements to the single pose example:
onnx
dependency if not neededonnxruntime
--runtime
argument)Maybe it makes sense to extract the methods and libraries into a util module to not have duplicate code in the single- and multi-pose example. For now, this PR already helps to run the models on the CPU (and MacOS / Windows) as well.