The BinaryAI / CodeCMR experiment consists of two steps. The first is an IDA Pro plugin that takes as input a JSON specifying which functions to consider for the tests, and as output it produces intermediate results in pickle format. In the second step a neural network is trained to recognise similar functions. The output of the neural network is a vector representation for each function. We release only Part 1, since the authors of the CodeCMR paper trained and tested the model for us (more information in the [technical report](../../Additional technical information.pdf)).
Before running the IDA plugin, follow the list of requirements from the IDA_scripts README
- Input: the JSON file with the selected functions (
-j) and the output directory (-o). - Output: one pickle file per IDB. The file contains a serialized version of a NetworkX graph (with the extracted features) for each function analyzed in the IDB.
Notes:
- The plugin requires the IDA Pro decompiler license
- IDA Pro requires 32-bit IDA to decompile 32-bit binaries and 64-bit IDA for 64-bit binaries.
32-bit binaries need to be exported to
.idbformat, while 64-bit binaries require.i64format. - The path of the IDB files in the JSON in input must be relative to the
binary_function_similaritydirectory. The Python3 script converts the relative path into a full path to correctly load the IDB in IDA Pro.
Example: run the plugins over the functions selected for the Dataset-1-CodeCMR test (requires the .i64 IDBs for 64-bit binaries and .idb for 32-bit binaries)
cd IDA_CodeCMR
python3 cli_codeCMR.py -j ../../../DBs/Dataset-1-CodeCMR/features/selected_Dataset-1-CodeCMR.json -o Dataset-1-CodeCMRRun unit tests:
python3 -m unittest test_codeCMR.pyIDA_CodeCMR.py includes part of the code from https://github.com/binaryai/sdk/ which is licensed under GPL-3.0.