OBF-BCSD

RoBERTa-based Obfuscated Binary Code Similarity Detection

Requirements

pip install -r requirements.txt

Besides, you need to install the followings:

• python3 (tested on 3.8)
• IDA Pro (tested on 8.2)

Testing

If you only want to test without preprocessing and training:

1. Download the RQ test dataset from https://zenodo.org/records/17119870.

2. Then move the files into the dataset directory:

mkdir dataset
mv RQ_test_dataset dataset/
mv dataset/RQ_test_dataset/* dataset/
rmdir dataset/RQ_test_dataset

3. Run the evaluation:

python eval.py

Dataset Preprocessing

You can download the binary datasets (ollvm.tar.xz and tigress.tar.xz) from

https://zenodo.org/records/17119870 and use them to perform the preprocessing steps for training and testing.

By default, put ollvm and tigress under the /data directory.

Step 1 — Generate initial dataset

python make_dataset.py --dataset_name ollvm
python make_dataset.py --dataset_name tigress

Step 2 — Generate tokenizer dataset

python make_tokenizer_dataset.py

Step 3 — Generate pretraining dataset

python make_pretrain_dataset.py --dataset_name ollvm
python make_pretrain_dataset.py --dataset_name tigress

Step 4 — Generate finetuning dataset

python make_finetune_dataset.py --dataset_name ollvm
python make_finetune_dataset.py --dataset_name tigress

---

Pretraining

python pretrain.py --dataset_name tigress
python pretrain.py --dataset_name ollvm

Finetuning

python finetune.py --dataset_name tigress 
python finetune.py --dataset_name ollvm

Finetuning Dataset Validation

python val_finetunedata.py --dataset_name tigress
python val_finetunedata.py --dataset_name ollvm

Create RQ Test Dataset

python make_rq_test_data.py

Evaluation

python eval.py