using cv:dnn caffe to load and inference darkscnn models with ros1 integration. It doesn't need caffe modules if you compile CV with CUDA support.

terminate called after throwing an instance of 'cv::Exception'
  what():  OpenCV(4.5.5) /home/s95huang/Downloads/opencv-4.5.5/modules/dnn/src/dnn.cpp:3231: error: (-215:Assertion failed) biasLayerData->outputBlobsWrappers.size() == 1 in function 'fuseLayers'

• looks like model error

  • The output should be 1 X 13 X img_size.
  • The cv.dnn deployment method causes dimmension mismatch.
  • Aborting further development

• Apollo deployed this caffe model by using TensorRT API to replace some components with equivalent CUDA ops.

  • For example, slice layer from caffe is rewritten using CUDA layer.

• Models are obtained from Apollo's Github repo.

  • Please refer to their website for licenses etc.

https://docs.opencv.org/4.5.5/d5/de7/tutorial_dnn_googlenet.html

https://github.com/ApolloAuto/apollo