Seeing Clearly, Answering Incorrectly: A Multimodal Robustness Benchmark for Evaluating MLLMs on Leading Questions

Overview

MMR provides a comprehensive suite to evaluate the understanding capabilities of Multimodal Large Language Models (MLLMs) and their robustness when handling negative questions after correctly interpreting visual content. The MMR benchmark includes:

Multimodal Robustness (MMR) Benchmark and Targeted Evaluation Metrics:
- Comprising 12 categories of paired positive and negative questions.
- Each question is meticulously annotated by experts to ensure scientific validity and accuracy.
Specially Designed Training Set:
- Contains paired positive and negative visual question-answer samples to enhance robustness.
Combined Dataset and Models:
- The new dataset merges the proposed dataset with existing ones.
- Trained models include Bunny-MMR-3B, Bunny-MMR-4B, and Bunny-MMR-8B.

In this repository, we provide Bunny-MMR-4B, which is built upon SigLIP and Phi-3-Mini-4K-Instruct. More details about this model can be found in GitHub.

Key Features

Rigorous Testing:
- Extensive testing on leading MLLMs shows that while these models can correctly interpret visual content, they exhibit significant vulnerabilities when faced with leading questions.
Enhanced Robustness:
- The targeted training significantly improves the MLLMs' ability to handle negative questions effectively.

Quickstart

Here we show a code snippet to show you how to use the model with transformers.

Before running the snippet, you need to install the following dependencies:

pip install torch transformers accelerate pillow

import torch
import transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
from PIL import Image
import warnings

# disable some warnings
transformers.logging.set_verbosity_error()
transformers.logging.disable_progress_bar()
warnings.filterwarnings('ignore')

# set device
torch.set_default_device('cpu')  # or 'cuda'

# create model
model = AutoModelForCausalLM.from_pretrained(
    'AI4VR/Bunny-MMR-4B',
    torch_dtype=torch.float16,
    device_map='auto',
    trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained(
    'AI4VR/Bunny-MMR-4B',
    trust_remote_code=True)

# text prompt
prompt = 'text prompt'
text = f"A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions. USER: <image>\n{prompt} ASSISTANT:"
text_chunks = [tokenizer(chunk).input_ids for chunk in text.split('<image>')]
input_ids = torch.tensor(text_chunks[0] + [-200] + text_chunks[1][1:], dtype=torch.long).unsqueeze(0)

# image, sample images can be found in images folder
image = Image.open('path/to/image')
image_tensor = model.process_images([image], model.config).to(dtype=model.dtype)

# generate
output_ids = model.generate(
    input_ids,
    images=image_tensor,
    max_new_tokens=100,
    use_cache=True)[0]

print(tokenizer.decode(output_ids[input_ids.shape[1]:], skip_special_tokens=True).strip())

Citation

If you find this repository helpful, please cite the paper below.

@misc{liu2024seeing,
    title={Seeing Clearly, Answering Incorrectly: A Multimodal Robustness Benchmark for Evaluating MLLMs on Leading Questions},
    author={Yexin Liu and Zhengyang Liang and Yueze Wang and Muyang He and Jian Li and Bo Zhao},
    year={2024},
    eprint={2406.10638},
    archivePrefix={arXiv},
}

License

The project employs specific datasets and checkpoints that are governed by their original licenses. Users must adhere to all terms and conditions outlined in these licenses. The checkpoints are restricted to uses that comply with the license agreements of Bunny, LLaMA 3, Phi-2, Phi-3, and GPT-4. The dataset is provided under the CC-BY-4.0 license.