video-reasoning
physical-commonsenseinteractive-video-reasoning
Dataset release
Videos, trajectories, manifests, and evaluation records are hosted at:
https://huggingface.co/datasets/sjmathy/interactive-video-reasoning
Expected layout:
data/
balance_rod/{task_bank,probe_bank}/
pushbar_finishline/{task_bank,probe_bank}/
slide_block/{task,probe}/
notched_launcher/shards/{task,probe}/
Domain
Task bank
Probe bank
Held-out rule
Balance-Rod
500 episodes
500 episodes
pose groups 80-99
Push-Bar
500 episodes
500 episodes
pose… See the full description on the dataset page: https://huggingface.co/datasets/sjmathy/interactive-video-reasoning.morse-500
MORSE-500 Benchmark
🔥 News
May 15, 2025: We release MORSE-500, 500 programmatically generated videos across six reasoning categories: abstract, mathematical, physical, planning, spatial, and temporal, to stress-test multimodal reasoning. Frontier models including OpenAI o3 and Gemini 2.5 Pro score lower than… See the full description on the dataset page: https://huggingface.co/datasets/video-reasoning/morse-500.morse-500-view
MORSE-500 Benchmark
The viewing version of MORSE-500 Benchmark, which allows you to view the video in the webpage directly.
Dataset Structure
test/: Contains all MP4 video files
test/metadata.csv: Contains the dataset metadata, including video_path, query, ground_truth, question_text, and main_category
video-latent-reasoningevidenceqa-core
EvidenceQA-Core
EvidenceQA-Core is a video reasoning dataset for evaluating whether a model can answer questions with explicit visual evidence. It contains temporal QA samples and spatial grounding samples derived from video understanding sources.
Dataset Contents
The release archive contains the complete local public_dataset/ layout:
public_dataset/
metadata.json
train.jsonl
validation.jsonl
test.jsonl
subsets/
balanced-v1.json
media/
videos/… See the full description on the dataset page: https://huggingface.co/datasets/evidence-video-reasoning/evidenceqa-core.