csp-cpp

Evaluating the Convolutional Social Pooling (CSP) work using the PyTorch C++ frontend

Installing

You can create an Anaconda environment, install CMake and the C++ distributions of PyTorch as follows:

1- Download the C++ distributions of PyTorch (LibTorch) ZIP archive:

wget https://download.pytorch.org/libtorch/nightly/cpu/libtorch-shared-with-deps-latest.zip
unzip libtorch-shared-with-deps-latest.zip

Note that the above link has CPU-only LibTorch.

2- Create and activate a new conda environment:

conda create --name cpp-pytorch
conda activate cpp-pytorch

3- Install CMake:

conda install -c anaconda cmake 

CMake is the recommended build system and will be well supported into the future.

Evaluating CSP using C++

This repository includes a 'eval.cpp' script that loads a traced model and evaluates it on a given set of inputs. The model is already traced and saved at '/traced-models' and a 'CMakeLists.txt' file is provide to build the application. Follow these steps to test the code:

1- Clone the repository to your local machine.

2- Run the following commands to build the application from within the cloned repository folder:

mkdir build
cd build
cmake -DCMAKE_PREFIX_PATH=/absolute/path/to/libtorch ..
cmake --build . --config Release

where /absolute/path/to/libtorch should be the ABSOLUTE path to the unzipped LibTorch distribution.

3- Execute the resulting binary found in the build folder:

./csp-cpp "../traced-models/traced_net_model.pt"

References

1- PyTorch documentation

https://pytorch.org/cppdocs/installing.html

https://pytorch.org/tutorials/advanced/cpp_export.html#:~:text=Step%201%3A%20Converting%20Your%20PyTorch,by%20the%20Torch%20Script%20compiler.

2- CSP Code

https://github.com/nachiket92/conv-social-pooling