Regarding the issue of C++ project deployment

[hefan14]

I noticed that PatchCore saves its outputs as a .pkl file (for model parameters) and a .faiss file (for the faiss index). While these files can be loaded in Python for inference, they do not appear to be directly usable for deployment in a C++ environment.

I would like to inquire whether you have considered the challenges of engineering deployment in your subsequent work. Specifically, are there plans or existing methods to export the PatchCore model to standardized formats like TorchScript or ONNX, which would facilitate cross-platform deployment (e.g., in C++ applications)?

Any insights or references on this matter would be greatly appreciated.

[dungdo123]

The Patchcore method is not a model but the memory bank approach. You should build the faiss C++ and use it for searching in memory bank during inference.