how to convet the model to onnx about mobilesam HOT 9 CLOSED

@qnmts Hi, please check this pull request.

from mobilesam.

@qnmts Hi, please check this pull request.

thank you !!!

from mobilesam.

@jnulzl This is my way of switching, but there are a lot of caveats,and the accuracy of ONNX's inference is not as high as before. Do you have this situation

Have you figured out the reason? If so, please kindly let us know. Thanks~~

from mobilesam.

@qnmts

python scripts/export_onnx_model.py --checkpoint weights/mobile_sam.pt --model-type vit_t --opset 11 --output mobile_sam_onnx_opset11.onnx

from mobilesam.

@jnulzl thank you， but no the vit_t type

from mobilesam.

@jnulzl This is my way of switching, but there are a lot of caveats,and the accuracy of ONNX's inference is not as high as before. Do you have this situation

from mobilesam.

@qnmts I didn't meet your error. You can try to update your code to the latest and try again.

from mobilesam.

@qnmts I didn't meet your error. You can try to update your code to the latest and try again.
ok

from mobilesam.

Hi @qnmts , we will close the issue.
Please let us know if there is performance degradation in exported ONNX model.

Thank you.

from mobilesam.