[Enhancement] Support RTMDet-Ins

[lzhangzz]

resolves #1849

[hanrui1sensetime]

Seems failed on python demo when the bbox is out of the picture because the mask shape and the mask_img shape are different.

FYI:
The model you can use rtmdet-ins-tiny-ort.
And the command is

python object_detection.py cpu ../../rtmdet-ins-tiny-ort/ ../../../mmdeploy/demo/resources/det.jpg

Then the error log is:

cv2.error: OpenCV(4.7.0) /io/opencv/modules/core/src/arithm.cpp:212: error: (-209:Sizes of input arguments do not match) The operation is neither 'array op array' (where arrays have the same size and type), nor 'array op scalar', nor 'scalar op array' in function 'binary_op'

The cause is:

debugging top: 104, top+mask.shape[0]: 141, left: 608, left+mask.shape[1]: 641
debugging mask: (37, 33), mask_img: (37, 32)

And the bug is at here

[lvhan028]

Seems failed on python demo when the bbox is out of the picture because the mask shape and the mask_img shape are different.

FYI: The model you can use rtmdet-ins-tiny-ort. And the command is

python object_detection.py cpu ../../rtmdet-ins-tiny-ort/ ../../../mmdeploy/demo/resources/det.jpg

Then the error log is:

cv2.error: OpenCV(4.7.0) /io/opencv/modules/core/src/arithm.cpp:212: error: (-209:Sizes of input arguments do not match) The operation is neither 'array op array' (where arrays have the same size and type), nor 'array op scalar', nor 'scalar op array' in function 'binary_op'

The cause is:

debugging top: 104, top+mask.shape[0]: 141, left: 608, left+mask.shape[1]: 641
debugging mask: (37, 33), mask_img: (37, 32)

And the bug is at here

Can you share the test image?

[hanrui1sensetime]

Seems failed on python demo when the bbox is out of the picture because the mask shape and the mask_img shape are different.
FYI: The model you can use rtmdet-ins-tiny-ort. And the command is

python object_detection.py cpu ../../rtmdet-ins-tiny-ort/ ../../../mmdeploy/demo/resources/det.jpg

Then the error log is:

cv2.error: OpenCV(4.7.0) /io/opencv/modules/core/src/arithm.cpp:212: error: (-209:Sizes of input arguments do not match) The operation is neither 'array op array' (where arrays have the same size and type), nor 'array op scalar', nor 'scalar op array' in function 'binary_op'

The cause is:

debugging top: 104, top+mask.shape[0]: 141, left: 608, left+mask.shape[1]: 641
debugging mask: (37, 33), mask_img: (37, 32)

And the bug is at here

Can you share the test image?

det.jpg

[hanrui1sensetime]

LGTM.

[lvhan028]

Seems failed on python demo when the bbox is out of the picture because the mask shape and the mask_img shape are different.
FYI: The model you can use rtmdet-ins-tiny-ort. And the command is

python object_detection.py cpu ../../rtmdet-ins-tiny-ort/ ../../../mmdeploy/demo/resources/det.jpg

Then the error log is:

cv2.error: OpenCV(4.7.0) /io/opencv/modules/core/src/arithm.cpp:212: error: (-209:Sizes of input arguments do not match) The operation is neither 'array op array' (where arrays have the same size and type), nor 'array op scalar', nor 'scalar op array' in function 'binary_op'

The cause is:

debugging top: 104, top+mask.shape[0]: 141, left: 608, left+mask.shape[1]: 641
debugging mask: (37, 33), mask_img: (37, 32)

And the bug is at here

Can you share the test image?

det.jpg

I didn't reproduce your result. Was it still an issue at your side?