This is c++ implementation to load a torchscript object detector.
The detector should take a tensor (1 x Channel(rgb) x Height x Width) as input and output tuple of 3 tensors (boudning box, class id, confidence score)
$ mkdir build
$ cd build
build$ cmake -DCMAKE_PREFIX_PATH=<libtorch_path> ..
build$ cmake --build . --config Release
build$ ./main -m <model_path> -i <image_path>
- Make docker file
- Video Detection