TorchServe KeyError

[austinmw]

Hi, I used MMDetection 3.0.0rc2 to train a Deformable DETR model and a Faster RCNN model, and then I used the mmdet2torchserve tool to create two model.mar. I tried testing local deployment with my own TorchServe Dockerfile, but I get errors with both of the TorchServe archives:

KeyError: "class DeformableDETR in mmdet/models/detectors/deformable_detr.py: 'DetDataPreprocessor is not in the model registry. Please check whether the value of DetDataPreprocessor is correct or it was registered as expected.

and

KeyError: "class FasterRCNN in mmdet/models/detectors/faster_rcnn.py: 'DetDataPreprocessor is not in the model registry. Please check whether the value of DetDataPreprocessor is correct or it was registered as expected.

My Dockerfile looks like this:

# Use Sagemaker PyTorch inference container as base image
ARG REGION
FROM 763104351884.dkr.ecr.${REGION}.amazonaws.com/pytorch-inference:1.12.1-gpu-py38-cu113-ubuntu20.04-sagemaker

# Install MMEngine, MMCV, MMDetection
WORKDIR /opt/ml/code
RUN pip install cython nvgpu && \
    pip install -U pip openmim torch-model-archiver && \
    pip install git+https://github.com/open-mmlab/[email protected] && \
    mim install "mmcv>=2.0.0rc2" -f https://download.openmmlab.com/mmcv/dist/cu113/torch1.12.0/index.html && \
    pip install git+https://github.com/open-mmlab/[email protected] && \
    rm -rf /var/lib/apt/lists/*

COPY config.properties /home/model-server/config.properties

WORKDIR /
ENTRYPOINT ["python", "/usr/local/bin/dockerd-entrypoint.py"]
CMD ["torchserve", "--start", "--ts-config", "/home/model-server/config.properties", "--model-store", "/opt/ml/model"]

I placed the model.mars into /opt/ml/model.

Anyone know how to resolve this KeyError? Is mmdet_handler.py compatible with v3.0.0rc2?