Paper page - VideoKR: Towards Knowledge- and Reasoning-Intensive Video Understanding

Source: https://huggingface.co/papers/2606.05259

Abstract

VideoKR presents a large-scale video reasoning dataset and benchmark designed to enhance knowledge-intensive video understanding through expert-domain content and human-in-the-loop example generation.

We introduce VideoKR, the firstlarge-scale training corpusspecifically designed to strengthen knowledge- and reasoning-intensive video understanding. It comprises 315Kvideo reasoningexamples over 145K newly collected, CC-licensed, expert-domain videos. We develop ahuman-in-the-loop,skill-oriented example generationpipeline that targets progressively deepervideo reasoningcapabilities while ensuring the difficulty, diversity, and reliability of both the examples and theirCoT rationales. We also curate VideoKR-Eval, a newexpert-annotated benchmarkwhere questions require genuine video understanding and knowledge-intensive reasoning rather than textual shortcuts. Our experiments show that, under a standard SFTrightarrowGRPO pipeline, models post-trained on VideoKR outperform priorpost-training approacheson knowledge-intensivevideo reasoningwhile remaining competitive on generalvideo reasoning, highlighting data design as a key driver of progress invideo reasoning. We further conduct comprehensive ablations to isolate the contributions of VideoKR, providing actionable insights for future work.

View arXiv pageView PDFGitHub9Add to collection

Community

Paper submitter

about 4 hours ago

VideoKR presents a large-scale video reasoning dataset and benchmark designed to enhance knowledge-intensive video understanding through expert-domain content and human-in-the-loop example generation.

Upload images, audio, and videos by dragging in the text input, pasting, orclicking here.

Tap or paste here to upload images

Get this paper in your agent:

hf papers read 2606\.05259

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper4

#### minuzero/VideoKR-Qwen3-VL-8B Video-Text-to-Text• 9B• Updatedabout 4 hours ago • 37 • 1 #### minuzero/VideoKR-Qwen2.5-VL-7B-SFT Video-Text-to-Text• 849k• Updatedabout 4 hours ago • 34 #### minuzero/VideoKR-Qwen3-VL-8B-SFT Video-Text-to-Text• 770k• Updatedabout 4 hours ago • 29 #### minuzero/VideoKR-Qwen2.5-VL-7B Video-Text-to-Text• 8B• Updatedabout 4 hours ago • 40

Datasets citing this paper2

#### minuzero/VideoKR-Eval Viewer• Updatedabout 4 hours ago • 2k • 741 #### minuzero/VideoKR-Train Viewer• Updatedabout 4 hours ago • 114k • 118

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2606.05259 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.