Hugging Face's logo

Datasets:
Paul691
/
X2I2

Tasks:
Image-to-Image
Text-to-Image
Any-to-Any
Languages:
English
Chinese
Size:
1M<n<10M
ArXiv:
License:
Dataset card Files
xet
 Community
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/webdataset/webdataset.py", line 81, in _split_generators
                  first_examples = list(islice(pipeline, self.NUM_EXAMPLES_FOR_FEATURES_INFERENCE))
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 32, in _get_pipeline_from_tar
                  fs: fsspec.AbstractFileSystem = fsspec.filesystem("memory")
                                                  ~~~~~~~~~~~~~~~~~^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/fsspec/registry.py", line 302, in filesystem
                  cls = get_filesystem_class(protocol)
                File "/usr/local/lib/python3.14/site-packages/fsspec/registry.py", line 239, in get_filesystem_class
                  raise ValueError(f"Protocol not known: {protocol}")
              ValueError: Protocol not known: memory
              
              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 71, 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.

project page arxiv demo demo model model model

X2I2 Dataset

  • 2025-08-17: jsons/inpaint_edit/ and images/inpaint_edit/edit_pf_one/ are being fixed, please do not download.
  • 2025-07-15: jsons/reflect/reflect.jsonl has been fixed and updated.
  • 2025-07-05: X2I2 are available now.

X2I2-video-editing 

# meta file (en): jsons/video_edit/edit_mv.jsonl
# meta file (zh): jsons/video_edit/edit_mv_zh.jsonl
# images:
cd images/video_edit/edit_mv_0 && cat edit_mv_0.tar.gz.part_* > edit_mv_0.tar.gz && tar -xzvf edit_mv_0.tar.gz
cd images/video_edit/edit_mv_1 && cat edit_mv_1.tar.gz.part_* > edit_mv_1.tar.gz && tar -xzvf edit_mv_1.tar.gz
...
cd images/video_edit/edit_mv_10 && cat edit_mv_10.tar.gz.part_* > edit_mv_10.tar.gz && tar -xzvf edit_mv_10.tar.gz

X2I2-inpaint-editing (Fixing the issue now. Do not download or use it!) 

# meta file (en): jsons/inpaint_edit/inpaint_edit.jsonl
# meta file (zh): jsons/inpaint_edit/inpaint_edit_zh.jsonl
# images:
cd images/inpaint_edit/edit_pf_obj && cat edit_pf_obj.tar.gz.part_* > edit_pf_obj.tar.gz && tar -xzvf edit_pf_obj.tar.gz
cd images/inpaint_edit/edit_pf_one && cat edit_pf_one.tar.gz.part_* > edit_pf_one.tar.gz && tar -xzvf edit_pf_one.tar.gz
cd images/inpaint_edit/icedit_mv && cat icedit_mv.tar.gz.part_* > icedit_mv.tar.gz && tar -xzvf icedit_mv.tar.gz

X2I2-in-context-generation 

# meta file (en): jsons/video_icgen/video_icgen.jsonl
# meta file (zh): jsons/video_icgen/video_icgen_zh.jsonl
# images:
cd images/video_icgen/icgen_mv_0 && cat icgen_mv_0.tar.gz.part_* > icgen_mv_0.tar.gz && tar -xzvf icgen_mv_0.tar.gz
cd images/video_icgen/icgen_mv_1 && cat icgen_mv_1.tar.gz.part_* > icgen_mv_1.tar.gz && tar -xzvf icgen_mv_1.tar.gz

X2I2-in-context-editing 

# meta file (en): jsons/video_icedit/video_icedit.jsonl
# meta file (zh): jsons/video_icedit/video_icedit_zh.jsonl
# images:
cd images/video_icedit/edit_ip && cat edit_ip.tar.gz.part_* > edit_ip.tar.gz && tar -xzvf edit_ip.tar.gz

X2I2-video-interleave 

# meta file (en): jsons/video_interleave/video_interleave.jsonl
# meta file (zh): jsons/video_interleave/video_interleave_zh.jsonl
# images:
cd images/video_interleave/x_mv && cat x_mv.tar.gz.part_* > x_mv.tar.gz && tar -xzvf x_mv.tar.gz

X2I2-reflection 

# meta file (en): jsons/reflect/reflect.jsonl
# images:
cd images/reflect/reflect && cat reflect.tar.gz.part_* > reflect.tar.gz && tar -xzvf reflect.tar.gz

Data format for reflection data:

"input_images" means the image generated by our model, while "output_instruction" is the reflection data for the corresponding input image. When "output_instruction[i]" is null, the corresponding image "input_images[i]" has no error so there is no reflection prompt. "output_image" is the ground truth for input instruction.

Here is some code for processing the data:

def process_reflection_example(self, example):

        input_images = example['input_images']
        output_image = example['output_image']
        output_instruction = example['output_instruction']

        user_instruction = example[example["used_instruction"]]
        prefix = f"<|im_start|>system\nYou are a helpful assistant capable of generating high-quality images based on user's descriptions. You will repeatedly evaluate whether the generated images meet the user's requirements, and if they do not, you will modify and regenerate the images until they fully align with the user's instructions.<|im_end|>\n<|im_start|>user\n{user_instruction}<|im_end|>\n<|im_start|>assistant\n"

        output_str = ""
        final_input_images = []
        for idx, (img, prompt) in enumerate(zip(input_images, output_instruction)):
            if random.random() < 0.1:
                prompt = None

            if prompt is None:
                final_input_images.append(output_image)
                output_str += "<|vision_start|><|image_pad|><|vision_end|>"
                answer = "The generated images meet the user's requirements, so there is no need to continue generating.<|im_end|>"
                break
            else:
                prompt = prompt.replace("\n", " ")
                prompt = normalize_whitespace(prompt)
                final_input_images.append(img)
                output_str += "<|vision_start|><|image_pad|><|vision_end|>"
                if idx == len(output_instruction) - 1:
                    answer = prompt + "\n"
                else:
                    output_str += prompt + "\n"
            
        instruction = prefix + output_str
        return instruction, answer, final_input_images, output_image
Downloads last month
50
Total file size:
3.9 TB

Paper for Paul691/X2I2