PICABench_results / README.md
Andrew613's picture
Upload PhysEditBenchV7 clean version: 15 models, 984 samples, 5000 QA pairs
4dce774 verified
metadata
dataset_info:
  features:
    - name: sample_id
      dtype: string
    - name: physics_category
      dtype: string
    - name: physics_law
      dtype: string
    - name: edit_operation
      dtype: string
    - name: difficulty
      dtype: string
    - name: author
      dtype: string
    - name: edit_instruction
      dtype: string
    - name: edit_area
      list:
        - name: height
          dtype: float64
        - name: id
          dtype: string
        - name: order
          dtype: int64
        - name: width
          dtype: float64
        - name: x
          dtype: float64
        - name: 'y'
          dtype: float64
    - name: keypoints
      struct:
        - name: Editing Completion
          dtype: string
        - name: Physics Consistency
          dtype: string
    - name: annotated_qa_pairs
      list:
        - name: answer
          dtype: string
        - name: box
          struct:
            - name: height
              dtype: float64
            - name: id
              dtype: string
            - name: order
              dtype: int64
            - name: width
              dtype: float64
            - name: x
              dtype: float64
            - name: 'y'
              dtype: float64
        - name: question
          dtype: string
    - name: output_path_GPT_Image_Edits
      dtype: string
    - name: output_path_Seedream_Edits
      dtype: string
    - name: output_path_Gemini_Edits
      dtype: string
    - name: output_path_qwenimg
      dtype: string
    - name: output_path_005_kontext_8k
      dtype: string
    - name: output_path_005_kontext_12k
      dtype: string
    - name: output_path_flux_kontext
      dtype: string
    - name: output_path_bagel_think
      dtype: string
    - name: output_path_bagel_nothink
      dtype: string
    - name: output_path_stepedit
      dtype: string
    - name: output_path_009_kontext_10k
      dtype: string
    - name: output_path_hidream_e1
      dtype: string
    - name: output_path_omnigen2
      dtype: string
    - name: output_path_uniworld
      dtype: string
    - name: output_path_DIMOO
      dtype: string
    - name: input_image
      dtype: image
    - name: output_image_GPT_Image_Edits
      dtype: image
    - name: output_image_Seedream_Edits
      dtype: image
    - name: output_image_Gemini_Edits
      dtype: image
    - name: output_image_qwenimg
      dtype: image
    - name: output_image_005_kontext_8k
      dtype: image
    - name: output_image_005_kontext_12k
      dtype: image
    - name: output_image_flux_kontext
      dtype: image
    - name: output_image_bagel_think
      dtype: image
    - name: output_image_bagel_nothink
      dtype: image
    - name: output_image_stepedit
      dtype: image
    - name: output_image_009_kontext_10k
      dtype: image
    - name: output_image_hidream_e1
      dtype: image
    - name: output_image_omnigen2
      dtype: image
    - name: output_image_uniworld
      dtype: image
    - name: output_image_DIMOO
      dtype: image
  splits:
    - name: train
      num_bytes: 2772754845
      num_examples: 984
  download_size: 2765505799
  dataset_size: 2772754845
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*