WGO-Bench is a robot video benchmark dataset released by Macrodata Labs. It aims to evaluate the ability of visual language models to convert robot and first-person action videos into timestamped subtask annotations. This dataset primarily focuses on two tasks: boundary detection and subtask annotation. The annotation labels emphasize describing the complete action events and state changes visible in the video clips.

Dataset composition:

• It contains 100 video episodes, encompassing 743 key sub-tasks and 63 unique task instructions.
• The data sources are divided into three categories: HomER first-person videos (25 videos), RoboInter DROID robotic arm videos (50 videos), and RoboCOIN Galaxea R1 Lite head-mounted camera videos (25 videos).
• The data is stored in Parquet format, with video files (MP4 bytes) directly embedded in each line of data.

Data Fields:

• id: A stable, unique identifier for a video clip.
• video: Directly embedded MP4 format video binary data
• instruction: The high-level task instruction corresponding to this segment
• segments: A list of gold-labeled segments, each element containing start_sec (start time), end_sec (end time), and subtask (subtask description).
• metadata: Source-specific additional information in JSON format