reSIM

Code for the "ReSIM: Re-ranking Binary Similarity Embeddings to Improve Function Search Performance" paper

Build Docker image

To build the docker image using the provided Dockerfile run

docker build -t resafe-img .

Then start a container using the script:

./run_container.sh

Run function search experiment

We provide the pool extracted from BinCorp as well as pre-computed embeddings for the CLAP model. To run the function search experiments, use the script:

launch_experiments.py

setting the appropriate configuration settings in config.yaml

To reproduce the tables and plots of the paper using pre-computed results, use the notebook

reproduce_results.ipynb

and decompress the .tar.zst archives of BinCorp, MultiComp and BinPool datasets.

Dataset

Resources for running the experiments as well as the fine-tuned reSIM model are available here