Dataset Viewer
Duplicate
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code:   StreamingRowsError
Exception:    ValueError
Message:      Invalid string class label BAGEL
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 149, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 129, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 544, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 3052, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2572, in __iter__
                  example = _apply_feature_types_on_example(
                      example, self.features, token_per_repo_id=self.token_per_repo_id
                  )
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2489, in _apply_feature_types_on_example
                  encoded_example = features.encode_example(example)
                File "/usr/local/lib/python3.14/site-packages/datasets/features/features.py", line 2244, in encode_example
                  return encode_nested_example(self, example)
                File "/usr/local/lib/python3.14/site-packages/datasets/features/features.py", line 1518, in encode_nested_example
                  {k: encode_nested_example(schema[k], obj.get(k), level=level + 1) for k in schema}
                      ~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/features/features.py", line 1541, in encode_nested_example
                  return schema.encode_example(obj) if obj is not None else None
                         ~~~~~~~~~~~~~~~~~~~~~^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/features/features.py", line 1210, in encode_example
                  example_data = self.str2int(example_data)
                File "/usr/local/lib/python3.14/site-packages/datasets/features/features.py", line 1147, in str2int
                  output = [self._strval2int(value) for value in values]
                            ~~~~~~~~~~~~~~~~^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/features/features.py", line 1168, in _strval2int
                  raise ValueError(f"Invalid string class label {value}")
              ValueError: Invalid string class label BAGEL

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

MLLM-Edit

Hugging Face GitHub Paper
Official implementation of MLLM-Edit from our NeurIPS 2026 paper β€œMLLM-Edit: Benchmarking Image Forgery Detection and Localization under MLLM-based Editing”.
Zeqin Yu, Ye Tian, Jian Zhang, Jiangqun Ni, Lanyun Zhu, Alex Kot, Xudong Jiang

Overview

MLLM-Edit Pipeline

MLLM-Edit is a large-scale benchmark for image forgery detection and localization (IFDL) under MLLM-based editing, comprising over 240k tampered images from diverse natural and text image sources. It covers 18 MLLM-based editors (8 open-source and 10 closed-source) and three tampering types: insertion, removal, and replacement.

MLLM-Edit is constructed through three stages: real image–prompt preparation, tampered image generation, and semi-automatic tampering mask annotation.

Dataset

MLLM-Edit consists of two complementary versions covering natural and text images.

Version Model Source MLLM-based Editors Images
MLLM-Edit-v1 Open-source (8) Qwen-IEPlus, OmniGen-v2, BAGEL, FLUX.2 [klein] Base 9B, Step1X-Edit v1.2, LongCat-Image, GLM-Image, JoyAI-Image-Edit 20,026
MLLM-Edit-v1 Closed-source (10) HunyuanImage 3.0, Kling Image 3.0, Wan 2.6 Image, Vidu Q2, Seedream 4.5, Jimeng 4.6, GPT Image 2/1.5, Grok Imagine, Reve V1, Nano Banana Pro (Gemini 3 Pro) 24,260
MLLM-Edit-v2 Open-source (5) Qwen-IEPlus, FLUX.2 [klein] Base 9B, Step1X-Edit v1.2, LongCat-Image, JoyAI-Image-Edit 197,927

In our experiments, MLLM-Edit-v1 is primarily used for evaluation, while MLLM-Edit-v2 provides larger subsets for model training.

Each tampered image is paired with its corresponding original image and editing prompt. Pixel-level tampering masks are provided for samples with valid annotations.

Download: πŸ€— Hugging Face

Structure:

MLLM-Edit/
β”œβ”€β”€ MLLM-Edit-v1/
β”‚   β”œβ”€β”€ Natural/
β”‚   β”‚   └── {Model}/       # 18 editors: 8 open-source + 10 closed-source
β”‚   β”‚       β”œβ”€β”€ Real/      # Original images
β”‚   β”‚       β”œβ”€β”€ Tamper/    # Tampered images
β”‚   β”‚       β”œβ”€β”€ Mask/      # Corresponding pixel-level binary masks
β”‚   β”‚       └── prompt.csv # Editing prompts
β”‚   └── Text/
β”‚       └── {Model}/       
β”‚           β”œβ”€β”€ Real/
β”‚           β”œβ”€β”€ Tamper/
β”‚           β”œβ”€β”€ Mask/
β”‚           └── prompt.csv
β”‚
└── MLLM-Edit-v2/
    β”œβ”€β”€ Natural/
    β”‚   └── {Model}/       # 5 open-source editors
    β”‚       β”œβ”€β”€ Real/
    β”‚       β”œβ”€β”€ Tamper/
    β”‚       β”œβ”€β”€ Mask/
    β”‚       └── prompt.csv
    └── Text/
        └── {Model}/       
            β”œβ”€β”€ Real/
            β”œβ”€β”€ Tamper/
            β”œβ”€β”€ Mask/
            └── prompt.csv

Code

We provide the semi-automatic mask annotation pipeline used to construct MLLM-Edit, including candidate mask generation, a human annotation interface, and demonstration samples.

Available: πŸ’» Mask Annotation Pipeline

Contact

We welcome researchers to use MLLM-Edit and appreciate citations to our paper. If you have any questions or encounter any issues, please feel free to contact us at πŸ“§ kimjyu@foxmail.com.

✍️ Citation

If you find MLLM-Edit useful for your research, please consider citing our paper:

@inproceedings{yu2026mllmedit,
  title={MLLM-Edit: Benchmarking Image Forgery Detection and Localization under MLLM-based Editing},
  author={Yu, Zeqin and Tian, Ye and Zhang, Jian and Ni, Jiangqun and Zhu, Lanyun and Kot, Alex and Jiang, Xudong},
  booktitle={The Fortieth Annual Conference on Neural Information Processing Systems},
  year={2026}
}
Downloads last month
181