search

ApolloAuto / apollo

An open autonomous driving platform
Apache License 2.0
25.01k stars 9.67k forks source link

About the deep learning dataset training frames for apollo's industrial grade product environment. #12891

Open hongyi-zhao opened 3 years ago

hongyi-zhao commented 3 years ago

Hi,

I noticed that the caffe deep learning framework is used by apollo for its perception module, see following file list for more info:

./modules/perception/testdata/camera/lib/obstacle/detector/yolo/data/ut/caffe.model
./modules/perception/testdata/camera/lib/obstacle/detector/yolo/data/ut/caffe.pt
./modules/perception/testdata/camera/lib/obstacle/detector/yolo/data/caffe.model
./modules/perception/testdata/camera/lib/obstacle/detector/yolo/data/caffe.pt
./modules/perception/testdata/camera/lib/obstacle/detector/yolo/data/visold/caffe.model
./modules/perception/testdata/camera/lib/obstacle/detector/yolo/data/visold/caffe.pt
./modules/perception/testdata/camera/lib/obstacle/detector/diepsie/data/res34_tdcl/baidu_iter_80000.caffemodel
./modules/perception/testdata/camera/lib/obstacle/transformer/multicue/data/caffe.model
./modules/perception/testdata/camera/lib/obstacle/transformer/multicue/data/caffe_rgb.model
./modules/perception/testdata/camera/lib/obstacle/transformer/multicue/data/caffe.pt
./modules/perception/testdata/camera/lib/lane/detector/darkSCNN/data/darkSCNN/deploy.caffemodel
./modules/perception/testdata/camera/lib/lane/detector/denseline/data/denseline/deploy.caffemodel
./modules/perception/testdata/camera/lib/lane/detector/denseline/data/denseline/deploy.caffemodel.metadata
./modules/perception/testdata/camera/lib/lane/postprocessor/darkSCNN/data/darkSCNN/deploy.caffemodel
./modules/perception/testdata/camera/lib/lane/postprocessor/denseline/data/denseline/deploy.caffemodel
./modules/perception/testdata/camera/lib/lane/postprocessor/denseline/data/denseline/deploy.caffemodel.metadata
./modules/perception/inference/caffe/caffe_net_test.cc
./modules/perception/inference/caffe/caffe_net.h
./modules/perception/inference/caffe/BUILD_
./modules/perception/inference/caffe/caffe_net.cc
./modules/perception/inference/inference_test_data/yolo/caffe.pt
./modules/perception/inference/inference_test_data/yolo/caffe.caffemodel.metadata
./modules/perception/inference/inference_test_data/denseline_parser/deploy.caffemodel.metadata

But it seems that the above datasets/examples are mainly for testing purpose. OTOH, the MLP Deep Learning Model doc told the following dependencies used there:

Install Dependencies

Run the following commands to install the necessary dependencies:

    Install numpy: conda install numpy
    Install tensorflow: conda install tensorflow
    Install keras: conda install -c conda-forge keras
    Install h5py: conda install h5py
    Install protobuf: conda install -c conda-forge protobuf
    Install PyTorch: conda install -c pytorch pytorch

It seems that for industrial grade product environment, the tensorflow/keras/pytorch etc. should be used instead of caffe. Am I right?

Regards, HY

jeroldchen commented 3 years ago

@hongyi-zhao Thanks for your advice! However, we have removed caffe support since Apollo 6.0. For inference context, we mainly use LibTorch and TensorRT.