Error when export TRT yolov8 with non-square input shape

[chuong98]

Prerequisite

• [X] I have searched the existing and past issues but cannot get the expected help.
• [X] I have read the FAQ documentation but cannot get the expected help.
• [X] The bug has not been fixed in the latest version.

🐞 Describe the bug

It throws the error when using the dynamic-shape TRT or static but non-square input, see config below for detail

ERROR - /root/workspace/mmdeploy/mmdeploy/apis/core/pipeline_manager.py - pop_mp_output - 80 - `mmdeploy.apis.utils.utils.to_backend` with Call id: 1 failed. exit.

Deploy Config for Dynamic TRT

_base_ = ['mmyolo::deploy/base_dynamic.py']
backend_config = dict(
    type='tensorrt',
    common_config=dict(fp16_mode=True, max_workspace_size=1 << 40),
    model_inputs=[
        dict(
            input_shapes=dict(
                input=dict(
                    min_shape=[1, 3, 640, 640],
                    opt_shape=[1, 3, 640, 640],
                    max_shape=[32, 3, 640, 640])))
    ])
use_efficientnms = False  # whether to replace TRTBatchedNMS plugin with EfficientNMS plugin # noqa E501

or Static-shape TRT with non-square input

_base_ = ['mmyolo::deploy/base_static.py']
onnx_config = dict(input_shape=(640, 800))
backend_config = dict(
    type='tensorrt',
    common_config=dict(fp16_mode=True, max_workspace_size=1 << 40),
    model_inputs=[
        dict(
            input_shapes=dict(
                input=dict(
                    min_shape=[1, 3, 640, 800],
                    opt_shape=[1, 3, 640, 800],
                    max_shape=[1, 3, 640, 800])))
    ]
)
use_efficientnms = True  # whether to replace TRTBatchedNMS plugin with EfficientNMS plugin # noqa E501

Command.

cd /root/workspace/mmdeploy && python tools/deploy.py \
    ${DEPLOY_CFG_PATH} \
    ${MODEL_CFG_PATH} \
    ${MODEL_CHECKPOINT_PATH} \
    ${INPUT_IMG} \
    --test-img ${TEST_IMG} \
    --work-dir ${WORK_DIR} \
    --device ${DEVICE} \
    --log-level INFO \
    --dump-info

Environment

Docker

Additional information

It works fine when using static-shape of square input image, e.g (640,640) but have the error when:

• compiling TRT Static-shape but not using square input, for example (480x640)
• compiling TRT Dynamic-shape with square input or non-square input

[chuong98]

I check the output deploy.json and the batch_size=1, no matter what I changed in the config. Is this normal?

{
    "version": "1.2.0",
    "task": "Detector",
    "models": [
        {
            "name": "yolodetector",
            "net": "end2end.engine",
            "weights": "",
            "backend": "tensorrt",
            "precision": "FP16",
            "batch_size": 1,
            "dynamic_shape": true
        }
    ],
    "customs": []
}