[Docs] How to convert rtmpose3d to tensorRT #3245
Description
📚 The doc issue
I exported the RTMPose3D model to ONNX and got three outputs with shapes:
(1, 133, 576)
(1, 133, 768)
(1, 133, 576)
Mycode is
import torch
from mmpose.apis import init_model
pose_estimator = init_model(
'./configs/rtmw3d-l_8xb64_cocktail14-384x288.py',
'rtmw3d-l_8xb64_cocktail14-384x288-794dbc78_20240626.pth',
device='cpu')
input = torch.randn(1, 3, 384, 288)
data_samples = {}
torch.onnx.export(
pose_estimator,
(input, data_samples),
"model.onnx",
input_names=["input"],
output_names=["output"],
verbose=True
)
onnx_model = onnx.load("model.onnx")
onnx.checker.check_model(onnx_model)
for i in onnx_model.graph.input:
print("Input:", i.name, i.type)
for o in onnx_model.graph.output:
print("Output:", o.name, o.type)
Could you clarify the meaning of these three outputs? Also, which one should be used for 3D keypoint inference?
Suggest a potential alternative/fix
No response