JackyZhuo/OpenXEditBench_filtered

OpenXEdit Dataset Filtered

Processing Status

• —Latest chunk processed: 0
• —Items per chunk: 1,000
• —Total items kept: 741
• —Total items removed: 259
• —Filter rate: 25.90%
• —Last updated: 2025-09-07 08:16:45 UTC

Dataset Description

This dataset contains high-quality embodied editing image pairs filtered from the original OpenXEdit dataset using GPT-5. Only pairs where the target image successfully demonstrates completion of the edit instruction are retained. All original columns from the source dataset are preserved.

Source Dataset

• —Input: hshjerry0315/OpenXEditBench

Filtering Criteria

Items are kept only if:

1. 1.The target image clearly shows the successful completion of the edit instruction
2. 2.The final state in the target image is properly aligned with the requested action
3. 3.The transformation from source to target follows the instruction accurately
4. 4.Both images are of good quality and the instruction is clear

Fields

All columns from the original dataset are preserved, including:

• —src_img: Source image before editing
• —tar_img: Target image after editing
• —edit_instruction: The editing instruction that should be performed
• —Additional metadata and fields from the original dataset

Processing Configuration

• —Model: gpt-5-mini-2025-08-07
• —Chunk size: 1,000
• —Max concurrent API calls: 100

Usage

from datasets import load_dataset

# Load all chunks
dataset = load_dataset("JackyZhuo/OpenXEditBench_filtered")

# Load specific chunks
dataset = load_dataset(
    "JackyZhuo/OpenXEditBench_filtered",
    data_files=["data/chunk_000.parquet", "data/chunk_001.parquet"]
)