Hugging Face's logo

Datasets:
UFOTeamwork
/
UFO-Bench

Dataset card Files
xet
 Community
2
Dataset Viewer
The dataset viewer is not available for this subset.
Cannot get the split names for the config 'default' of the dataset.
Exception:    SplitsNotFoundError
Message:      The split names could not be parsed from the dataset config.
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
                  for split_generator in builder._split_generators(
                                         ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/folder_based_builder/folder_based_builder.py", line 246, in _split_generators
                  raise ValueError(
                      "`file_name`, `*_file_name`, `file_names` or `*_file_names` must be present as dictionary key in metadata files"
                  )
              ValueError: `file_name`, `*_file_name`, `file_names` or `*_file_names` must be present as dictionary key in metadata files
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 68, in compute_split_names_from_streaming_response
                  for split in get_dataset_split_names(
                               ~~~~~~~~~~~~~~~~~~~~~~~^
                      path=dataset,
                      ^^^^^^^^^^^^^
                      config_name=config,
                      ^^^^^^^^^^^^^^^^^^^
                      token=hf_token,
                      ^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
                  info = get_dataset_config_info(
                      path,
                  ...<6 lines>...
                      **config_kwargs,
                  )
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
                  raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
              datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.

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.

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

UFO-Bench

UFO-Bench is a benchmark for evaluating subject-driven and personalized image generation/editing systems under text-image conditioned scenarios.

The benchmark is designed to evaluate not only visual quality, but also whether a generated image faithfully preserves the identity, semantic attributes, and structural components of the reference subject while correctly following textual editing instructions.

In contrast to conventional subject-driven generation benchmarks, UFO-Bench places particular emphasis on complex editing instructions and conflicting text-image conditions, providing a more challenging and comprehensive evaluation setting for modern image generation and editing models.

Overview

Subject-driven image generation aims to generate images that preserve a specific subject from one or more reference images while following a textual prompt.

A successful model needs to satisfy multiple requirements simultaneously:

  • Subject consistency: preserve the identity and visual characteristics of the reference subject.
  • Attribute preservation: maintain important subject-level attributes such as color, shape, clothing, hairstyle, and other semantic properties.
  • Text instruction following: correctly execute the requested modifications.
  • Local editing capability: modify specific parts or attributes while preserving the rest of the subject.
  • Global editing capability: perform scene-level or subject-level transformations.
  • Complex condition interaction: correctly resolve interactions and potential conflicts between the reference image and textual instructions.

UFO-Bench is designed around these requirements.

Benchmark Statistics

Property UFO-Bench
Total cases 660
Subject categories 7
Editing paradigms 4
Reference images 86 unique subjects
Ground-truth annotations T_gt
Question annotations Question_list
Generated results generated/
Reference images reference/
Metadata metadata.jsonl

Subject Categories

UFO-Bench covers seven representative subject categories:

  1. Rigid Object
  2. Soft Object
  3. Human
  4. Full-body Character
  5. Animal
  6. Logo
  7. Scene

These categories cover both object-centric and scene-level subject-driven generation scenarios.

Editing Paradigms

The benchmark contains four major editing paradigms:

1. Non-Editing

The model is required to preserve the reference subject while generating the requested scene or context without explicitly modifying the subject.

This setting primarily evaluates:

  • subject identity preservation
  • visual consistency
  • text-image alignment

2. Local Editing

The instruction modifies a specific part, attribute, or local region of the subject.

Examples include:

  • changing clothing color
  • modifying hairstyle
  • changing an object component
  • modifying a specific visual attribute

This setting evaluates whether the model can perform fine-grained localized editing while preserving unrelated subject characteristics.

3. Global Editing

The instruction introduces a broader transformation affecting the subject or surrounding scene.

This evaluates the model's ability to perform large-scale semantic changes while maintaining the underlying subject identity.

4. Complex Editing

Complex Editing contains multiple interacting editing requirements and is designed to stress-test the model's ability to jointly satisfy multiple conditions.

These cases may involve:

  • multiple subject attributes
  • multiple editing operations
  • interactions between subject and scene
  • conflicting text-image conditions

Complex Editing is one of the key challenging components of UFO-Bench.

Conflicting Text-Image Conditions

A major feature of UFO-Bench is the inclusion of conflicting text-image conditions.

In these cases, the textual instruction may explicitly request a property that differs from the corresponding property observed in the reference image.

For example:

Reference image: a red object
Text instruction: a blue object

A capable model should understand that the textual instruction specifies an edit, rather than simply copying the original visual attribute.

This setting evaluates whether a model can correctly reason over the interaction between:

  • reference-image information
  • textual instructions
  • requested modifications

UFO-Bench contains a substantially larger proportion of such challenging cases than several existing subject-driven generation benchmarks.

Dataset Structure

The dataset is organized as follows:

UFO-Bench/
├── generated/
│   ├── ...
│   └── ...
│
├── Question_list/
│   ├── ...
│   └── ...
│
├── reference/
│   ├── ...
│   └── ...
│
├── T_gt/
│   ├── ...
│   └── ...
│
└── metadata.jsonl
Downloads last month
33
Total file size:
21.6 GB