AllentDan

Hi, @timothylimyl. Seems `No CUDA runtime is found` during building MMCV with TensorRT. Maybe you can refer to [mmdeploy](https://github.com/open-mmlab/mmdeploy) and its `Dockerfile` for some luck.

That seems to be the env variable should be specified in the Dockerfile. Try the methods from [mmdet issue 281](https://github.com/open-mmlab/mmdetection/issues/281) and envs in [mmdeploy dockerfile](https://github.com/open-mmlab/mmdeploy/blob/master/docker/GPU/Dockerfile) please.

> Hi @AllentDan , my previous dockerfile already has those env `ENV FORCE_CUDA="1"` and `ENV DEBIAN_FRONTEND=noninteractive` > > Edit: add my requirements.txt for completeness: > > ``` > --find-links https://download.pytorch.org/whl/torch_stable.html...

I mean: ```shell RUN sed -i '144,145d' setup.py && sed -i '142 i\ \ \ \ \ \ \ \ tensorrt_lib_path = "/usr/lib/x86_64-linux-gnu/"' setup.py ``` And make sure `RUN python...

Hi, please refer to the closed issue in MMOCR [here](https://github.com/open-mmlab/mmocr/issues/678). MMOCR recog models do not fully support dynamic batch inference because of `valid_ratios`.

May update the building way of the dockerfile as well.

Satrn accepts 3-channel inputs. Please use `configs/mmocr/text-recognition/text-recognition_tensorrt_static-32x32.py` instead.

Oh, please use satrn_small instead. 2GB is the limit size of ONNX protobuf.

The ONNX file should also be generated by the TensorRT configs other than ONNXRuntime configs.

Even if it is fine on another computer, that does not mean it is right. Like I said, if you want to use ONNXRuntime to do the inference, just use...