Skip to content

About

Code for FusionAudit: Pathway-Conditioned Robustness Auditing for Native Multimodal Models (NeurIPS 2026)

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

2 Commits

Folders and files

Repository files navigation

FusionAudit

FusionAudit measures how easily a vision-language model can be misled through the image alone. It runs five attacks on two visual pathways and reports, for each model, the share of correctly answered questions that the attacks break.

Code of the NeurIPS 2026 paper FusionAudit: Pathway-Conditioned Robustness Auditing for Native Multimodal Models. Project page

Attacks

Pathway Attack What it does
Pixel perturbation Vis-PGD Pushes the vision-encoder features away from those of the clean image.
End-PGD Lowers the probability of the correct answer.
MANTIS Raises a wrong answer over the correct one at the first token where the two differ.
Rendered text Typo-Evid Writes a wrong answer on the image as a text overlay.
Nat-Evid Edits the image so that a wrong answer appears as part of it, such as a chart label or a sign.

The pixel attacks are projected gradient descent on the RGB pixels of the original image, with an L∞ budget of 8/255, 40 steps of size 1/255 and a random start (seed 1).

Models

18 open-weight models, loaded from the Hugging Face Hub at pinned revisions (configs/models.json) and grouped by how deeply vision is fused into the language model:

Level Models
L1, alignment-era SmolVLM2-2.2B, BLIP2-OPT-2.7B, InstructBLIP-Vicuna-7B, LLaVA-1.5-7B
L2, shallow fusion Qwen-VL-Chat, Qwen2-VL-2B, Qwen2.5-VL-3B/7B, Gemma3-4B/12B, DeepSeek-VL-7B
L3, deep or early fusion Qwen3-VL-2B/4B/8B, Qwen3.5-2B/4B/9B, Gemma4-E4B

Installation

Python 3.11.

pip install -e .            # data preparation and summaries
pip install -e '.[main]'    # inference and attacks
Extra Use
main inference and attacks for 16 of the models (transformers 5.5.3)
legacy Qwen-VL-Chat and DeepSeek-VL-7B (transformers 4.32.0), in a separate environment
generation Nat-Evid image generation
planning recomputing the Nat-Evid wrong answers

See docs/ENVIRONMENTS.md. Gated models such as Gemma need a Hugging Face login (docs/WEIGHTS.md).

Data

Evaluation set. 1500 VQA questions, 500 each from ChartQA (train), DocVQA (validation) and MME (test). The repository holds an index with the dataset row and image hash of every question. The script downloads the questions and images from the Hugging Face Hub and checks them against the index:

python scripts/build_eval_set.py --index data/index.json \
    --out data/manifest.json --image-cache data/images

Typo-Evid images, rendered locally:

python scripts/render_typo_evid.py --manifest data/manifest.json --image-cache data/images \
    --out data/typo_evid --hashes data/typo_evid_hashes.json

Nat-Evid images, edited with the Gemini API. The wrong answer for each question is fixed in data/nat_evid/plans.jsonl. This is the only step that needs an API key.

pip install -e '.[generation]'
export GEMINI_API_KEY=...
python scripts/generate_nat_evid.py generate --manifest data/manifest.json \
    --image-cache data/images --out data/nat_evid

Running an audit

Run the clean pass first: the attacks use the questions the model answers correctly.

ARGS="--model Qwen/Qwen2.5-VL-3B-Instruct --manifest data/manifest.json"
ARGS="$ARGS --image-cache data/images --records runs/records"

python scripts/run_clean.py $ARGS
for a in vis_pgd end_pgd mantis; do
    python scripts/run_pixel_attacks.py $ARGS --attack $a
done
python scripts/run_rendered_eval.py $ARGS --condition typo_evid --images data/typo_evid
python scripts/run_rendered_eval.py $ARGS --condition nat_evid --images data/nat_evid/images

python scripts/summarize.py --manifest data/manifest.json \
    --records runs/records --out runs/results

Each run appends to runs/records/<model>.jsonl and resumes where it stopped. Options:

  • --limit N or --sample-ids FILE runs part of the set; pass --allow-incomplete to summarize.py for such runs.
  • --eps, --steps, --alpha and --seed change the pixel attacks.
  • --revision loads another commit of the model.

Outputs

summarize.py judges every answer again from the generated text and writes two files.

main_table.csv has one row per model:

Column Meaning
clean_acc share of questions answered correctly without an attack
vis_pgd ... nat_evid attack success rate: share of the correctly answered questions that the attack turns wrong
delta_mantis share broken by MANTIS and by neither Vis-PGD nor End-PGD
delta_nat_evid share broken by Nat-Evid and not by Typo-Evid
rhat0 share broken by at least one of Vis-PGD, End-PGD and Typo-Evid
rhat_plus share broken by at least one of the five attacks
delta_rhat_plus rhat_plus minus rhat0

level_summary.csv averages the same columns over the models of each level.

Released model outputs

data/records/ holds the answers of the 18 models on the clean and attacked images used in the paper. Summarizing them reproduces the paper's results without a GPU:

python scripts/summarize.py --manifest data/manifest.json --records data/records --out results

Citation

@inproceedings{zhang2026fusionaudit,
  title     = {FusionAudit: Pathway-Conditioned Robustness Auditing
               for Native Multimodal Models},
  author    = {Zhang, Jiaming and Li, Zherui and Guo, Siqi and
               Dong, Junhao and Wang, Kun and Lim, Wei Yang Bryan},
  booktitle = {Advances in Neural Information Processing Systems},
  year      = {2026}
}

Released under the MIT License.

About

Code for FusionAudit: Pathway-Conditioned Robustness Auditing for Native Multimodal Models (NeurIPS 2026)

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages