[Bug] Quantized RTMDet model does not detect objects/humans

[guntisuSICKAG]

Checklist

• I have searched related issues but cannot get the expected help.
• 2. I have read the FAQ documentation but cannot get the expected help.
• 3. The bug has not been fixed in the latest version.

Describe the bug

After I downloaded the model configs and checkpoints of rtmdet_small (rtmdet_s_8xb32-300e_coco_20220905_161602-387a891e) model, I converted the model to onnx backend using torch2onnx and then to tensorrt backend using onnx2tensorrt along with the calibration file generated using create_calib_input_data for quantization purposes. The resulting quantized model does not detect objects/humans. However, when I tested the quantized YoloX and DeTR models, they could detect humans pretty well.

Reproduction

1. Download the RTMDet model using "mim download mmdet --config rtmdet_s_8xb32-300e_coco_20220905_161602-387a891e --dest ." .
2. Generate calibration file using create_calib_input_data from mmdeploy.apis.calibration
3. Conversion to ONNX backend using torch2onnx from mmdeploy.apis
4. Conversion to TRT backend using onnx2tensorrt from mmdeploy.backend.tensorrt.onnx2tensorrt

Environment

Environment Info:

01/13 13:24:38 - mmengine - INFO - 

01/13 13:24:38 - mmengine - INFO - **********Environmental information**********
01/13 13:24:39 - mmengine - INFO - sys.platform: linux
01/13 13:24:39 - mmengine - INFO - Python: 3.8.10 (default, Nov  7 2024, 13:10:47) [GCC 9.4.0]
01/13 13:24:39 - mmengine - INFO - CUDA available: True
01/13 13:24:39 - mmengine - INFO - MUSA available: False
01/13 13:24:39 - mmengine - INFO - numpy_random_seed: 2147483648
01/13 13:24:39 - mmengine - INFO - GPU 0: Orin
01/13 13:24:39 - mmengine - INFO - CUDA_HOME: /usr/local/cuda
01/13 13:24:39 - mmengine - INFO - NVCC: Cuda compilation tools, release 11.4, V11.4.315
01/13 13:24:39 - mmengine - INFO - GCC: aarch64-linux-gnu-gcc (Ubuntu 9.4.0-1ubuntu1~20.04.2) 9.4.0
01/13 13:24:39 - mmengine - INFO - PyTorch: 1.11.0
01/13 13:24:39 - mmengine - INFO - PyTorch compiling details: PyTorch built with:
  - GCC 9.4
  - C++ Version: 201402
  - OpenMP 201511 (a.k.a. OpenMP 4.5)
  - LAPACK is enabled (usually provided by MKL)
  - NNPACK is enabled
  - CPU capability usage: NO AVX
  - CUDA Runtime 11.4
  - NVCC architecture flags: -gencode;arch=compute_72,code=sm_72;-gencode;arch=compute_87,code=sm_87
  - CuDNN 8.6
    - Built with CuDNN 8.3.2
  - Build settings: BLAS_INFO=open, BUILD_TYPE=Release, CUDA_VERSION=11.4, CUDNN_VERSION=8.3.2, CXX_COMPILER=/usr/bin/c++, CXX_FLAGS= -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOCUPTI -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -DEDGE_PROFILER_USE_KINETO -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wno-sign-compare -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-unused-local-typedefs -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Werror=cast-function-type -Wno-stringop-overflow, FORCE_FALLBACK_CUDA_MPI=1, LAPACK_INFO=open, TORCH_VERSION=1.11.0, USE_CUDA=ON, USE_CUDNN=ON, USE_EIGEN_FOR_BLAS=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=OFF, USE_MKLDNN=OFF, USE_MPI=ON, USE_NCCL=0, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, 

01/13 13:24:39 - mmengine - INFO - TorchVision: 0.12.0a0+9b5a3fe
01/13 13:24:39 - mmengine - INFO - OpenCV: 4.10.0
01/13 13:24:39 - mmengine - INFO - MMEngine: 0.10.5
01/13 13:24:39 - mmengine - INFO - MMCV: 2.1.0
01/13 13:24:39 - mmengine - INFO - MMCV Compiler: GCC 9.4
01/13 13:24:39 - mmengine - INFO - MMCV CUDA Compiler: 11.4
01/13 13:24:39 - mmengine - INFO - MMDeploy: 1.3.1+af7ac02
01/13 13:24:39 - mmengine - INFO - 

01/13 13:24:39 - mmengine - INFO - **********Backend information**********
01/13 13:24:39 - mmengine - INFO - tensorrt:    8.5.2.2
01/13 13:24:39 - mmengine - INFO - tensorrt custom ops: Available
01/13 13:24:39 - mmengine - INFO - ONNXRuntime: None
01/13 13:24:39 - mmengine - INFO - ONNXRuntime-gpu:     1.16.0
01/13 13:24:39 - mmengine - INFO - ONNXRuntime custom ops:      NotAvailable
01/13 13:24:39 - mmengine - INFO - pplnn:       None
01/13 13:24:39 - mmengine - INFO - ncnn:        None
01/13 13:24:39 - mmengine - INFO - snpe:        None
01/13 13:24:39 - mmengine - INFO - openvino:    None
01/13 13:24:39 - mmengine - INFO - torchscript: 1.11.0
01/13 13:24:39 - mmengine - INFO - torchscript custom ops:      NotAvailable
01/13 13:24:39 - mmengine - INFO - rknn-toolkit:        None
01/13 13:24:39 - mmengine - INFO - rknn-toolkit2:       None
01/13 13:24:39 - mmengine - INFO - ascend:      None
01/13 13:24:39 - mmengine - INFO - coreml:      None
01/13 13:24:39 - mmengine - INFO - tvm: None
01/13 13:24:39 - mmengine - INFO - vacc:        None
01/13 13:24:39 - mmengine - INFO - 

01/13 13:24:39 - mmengine - INFO - **********Codebase information**********
01/13 13:24:39 - mmengine - INFO - mmdet:       3.2.0
01/13 13:24:39 - mmengine - INFO - mmseg:       None
01/13 13:24:39 - mmengine - INFO - mmpretrain:  None
01/13 13:24:39 - mmengine - INFO - mmocr:       None
01/13 13:24:39 - mmengine - INFO - mmagic:      None
01/13 13:24:39 - mmengine - INFO - mmdet3d:     None
01/13 13:24:39 - mmengine - INFO - mmpose:      1.3.2
01/13 13:24:39 - mmengine - INFO - mmrotate:    None
01/13 13:24:39 - mmengine - INFO - mmaction:    None
01/13 13:24:39 - mmengine - INFO - mmrazor:     None
01/13 13:24:39 - mmengine - INFO - mmyolo:      None

Error traceback

No response