Datasetpublic

wangyueqian/Hum-Omni-Track2-Phase1

Track 2 Phase 1 Test Data This test set is part of the HumOmni 2026 competition. Competition website: https://humomni2026.github.io/ Overview The test set contains 500 video question-answering samples. Each sample consists of a natural language question about a short video clip, along with frames pre-extracted at 2 fps for convenience. Data Format Each sample is stored in its own directory under data/, named by the video ID (e.g.… See the full description on the dataset page: https://huggingface.co/datasets/wangyueqian/Hum-Omni-Track2-Phase1.

sourceHugging Faceupdated 4mo agoView on Hugging Face

0likes5downloads

Dataset Card

Track 2 Phase 1 Test Data

This test set is part of the HumOmni 2026 competition.

Competition website: https://humomni2026.github.io/

Overview

The test set contains 500 video question-answering samples. Each sample consists of a natural language question about a short video clip, along with frames pre-extracted at 2 fps for convenience.

Data Format

Each sample is stored in its own directory under data/, named by the video ID (e.g., data/OSfMU69X3C4.7/).

data/
└── <video_id>/
    ├── question.json   # the question for this sample
    ├── 0.5.jpg         # frame at 0.5 sec
    ├── 1.0.jpg         # frame at 1.0 sec
    ├── 1.5.jpg
    └── ...             # frames continue at 0.5 sec intervals

`question.json`

json

{
  "question_id": "OSfMU69X3C4.7.mp4",
  "question": "What kind of items are sold in the store?"
}

Field	Description
`question_id`	Unique identifier for the sample (video filename)
`question`	The question to be answered based on the video

Frames

Frames are extracted at 2 fps (one frame every 0.5 seconds), scaled to 480px width, and saved as JPEG named by their timestamp in seconds (0.5.jpg, 1.0.jpg, ...). The last frame corresponds to the end of the video clip.