# RTMW (OpenMMLab's rtmlib ONNX SDK files) as fp16 with float32 inputs and outputs, for onnxruntime-web on WebGPU.
#   python convert_rtmw.py rtmw-dw-l-m_simcc-cocktail14.onnx rtmw-m-256x192-fp16.onnx
#   python convert_rtmw.py rtmw-dw-x-l_simcc-cocktail14.onnx rtmw-l-384x288-fp16.onnx
# The sources: download.openmmlab.com/mmpose/v1/projects/rtmw/onnx_sdk/rtmw-dw-l-m_simcc-cocktail14_270e-256x192_20231122.zip
# and .../rtmw-dw-x-l_simcc-cocktail14_270e-384x288_20231122.zip (end2end.onnx in each).
import sys, onnx
from onnxruntime.transformers.float16 import convert_float_to_float16
from dedupe import shrink
m, _, _ = shrink(convert_float_to_float16(onnx.load(sys.argv[1]), keep_io_types=True))
onnx.checker.check_model(m)
onnx.save(m, sys.argv[2])