README.md

📝 torch_cpp

---

🎛 Dependencies

---

• CUDA Toolkits: if you use GPU.

• pytorch c++ API: the easiest way is to reuse build binary provided from pytorch official website. Here is the sample script to use the build binary. For convenience, this repo assumes torch c++ api is installed into /opt/libtorch

• other dependencies:

sudo apt-get install -y --no-install-recommends \
    libopencv-dev

🔨 How to Build

---

# build library
make default

# build examples
make apps -j`nproc`

# build gpu examples
make gpu_apps -j`nproc`

🏃 How to Run

SuperPoint and SuperGlue

---

Usage

• Script pre-trained superpoint and superglue weights

git submodule update --init --recursive
export ROOT_DIR="$(pwd)"
cd $ROOT_DIR/scripts/superglue/SuperGluePretrainedNetwork/
git apply ../jit_patch.patch
python3 -m pip install -r $ROOT_DIR/scripts/superglue/SuperGluePretrainedNetwork/requirements.txt
cd $ROOT_DIR
python3 $ROOT_DIR/scripts/superglue/jit_superglue_model.py
python3 $ROOT_DIR/scripts/superglue/jit_superpoint_model.py

• Test inference apps

./build/examples/match_images_superglue/match_images_superglue_app path/to/superpoint_model.pt path/to/superglue_model.pt ./data/images/VisionCS_0a.png ./data/images/VisionCS_0b.png

(back to top)

💎 References

---

• SuperGlue trained by magicleap
• SuperPoint SLAM
• Torch Tracing vs Torch Scripting

(back to top)