datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Video-R1-dataThis repository contains the data presented in Video-R1: Reinforcing Video Reasoning in MLLMs.
Code: https://github.com/tulerfeng/Video-R1
Video data folder: CLEVRER, LLaVA-Video-178K, NeXT-QA, PerceptionTest, STAR
Image data folder: Chart, General, Knowledge, Math, OCR, Spatial
Video-R1-COT-165k.json is for SFT cold start, and Video-R1-260k.json is for RL training.
Data Format in Video-R1-COT-165k:
{
"problem_id": 2,
"problem": "What appears on the screen in Russian during the… See the full description on the dataset page: https://huggingface.co/datasets/Video-R1/Video-R1-data.Embodied-R1.5-SFT-Dataset
Embodied-R1.5-SFT-Dataset
🌐 Project Page |
📄 arXiv |
💻 Code |
🧰 EmbodiedEvalKit |
🤗 Models & Datasets
🗓️ Update — 2026-08-20 (20260820). All 34 Stage 1 SFT JSON annotation files have been uploaded to sft_datasets_json/. The complete JSON ↔ image/video data mapping is documented in the Dataset composition table below.
⚠️ Partial release. This repository currently contains only a subset of the full Stage 1 SFT… See the full description on the dataset page: https://huggingface.co/datasets/IffYuan/Embodied-R1.5-SFT-Dataset.Mobile-R1
Dataset Card for Mobile-R1
Dataset Structure
images/: All the screenshots
data.jsonl: The trajectory data
Data Fields
All screenshots are stored in the images/ directory.
We describe the structure of a single trajectory entry from the file data.jsonl, which contains the full interaction trajectories and action history.
app_name: String. The name of the mobile application (e.g., "闲鱼" / Xianyu) where the task is performed.
trajectory_length: Integer. Number of… See the full description on the dataset page: https://huggingface.co/datasets/Mobile-R1/Mobile-R1.Mobile-R1
Dataset Card for Mobile-R1
Dataset Structure
images/: All the screenshots
data.jsonl: The trajectory data
Data Instances
Data Fields
All screenshots are stored in the images/ directory.
We describe the structure of a single trajectory entry from the file data.jsonl, which contains the full interaction trajectories and action history.
app_name: String. The name of the mobile application (e.g., "闲鱼" / Xianyu) where the task is performed.… See the full description on the dataset page: https://huggingface.co/datasets/PG23/Mobile-R1.book2skill-qc-r11Embodied-R1.5-RFT-Dataset
Embodied-R1.5-RFT-Dataset
🌐 Project Page |
📄 arXiv |
💻 Code |
🧰 EmbodiedEvalKit |
🤗 Models & Datasets
🗓️ Update — 2026-08-20 (20260820). All 28 Stage 2 RFT JSON annotation files have been uploaded to rft_datasets_json/. The complete JSON ↔ media archive mapping is documented in the Dataset composition table below.
⚠️ Partial release. This repository currently contains only a subset of the full Stage 2 RFT data… See the full description on the dataset page: https://huggingface.co/datasets/IffYuan/Embodied-R1.5-RFT-Dataset.im2vec-svg-stack-sample
im2vec-svg-stack-sample
Rendered (SVG, PNG) pairs for training Im2Vec,
a raster-logo-to-SVG model. This is a pre-rendered sample of
starvector/svg-stack
(2.17M rows total, scraped from permissively-licensed GitHub repos): each
.svg is paired with a same-stem .png rendered at 256x256 via
cairosvg, matching this project's training
pipeline (im2vec/data/render.py).
Why this dataset
Added alongside im2vec-svg-emoji
to give the model more data and more shape diversity… See the full description on the dataset page: https://huggingface.co/datasets/R1l3y-w/im2vec-svg-stack-sample.multimodal-open-r1-8k-verifiedMobile-R1
Dataset Card for Mobile-R1
Dataset Structure
images/: All the screenshots
data.jsonl: The trajectory data
Data Fields
All screenshots are stored in the images/ directory.
We describe the structure of a single trajectory entry from the file data.jsonl, which contains the full interaction trajectories and action history.
app_name: String. The name of the mobile application (e.g., "闲鱼" / Xianyu) where the task is performed.
trajectory_length:… See the full description on the dataset page: https://huggingface.co/datasets/huCamus/Mobile-R1.Robust-R1Robust-R1:Degradation-Aware Reasoning for Robust Visual Understanding
✅ Copyright Statement
Our dataset is built upon a subset of A-OKVQA (Schwenket al. 2022), comprising 10K samples for training and 1K for validation.
This statement serves to clarify that the copyright for original visual content is held by the provider.
If you wish to use original content itself, you must obtain permission from the respective copyright holders.
⭐️ Citation
If you find Robust-R1 useful… See the full description on the dataset page: https://huggingface.co/datasets/Jiaqi-hkust/Robust-R1.GEOQA_R1V_Train_8KProcessed from Geo170K, we filtered out the questions with the same (image, answer) pair.
Ocean_R1_collected_visual_dataVLM-R1This repository contains the dataset used in the paper VLM-R1: A Stable and Generalizable R1-style Large Vision-Language Model.
Code: https://github.com/om-ai-lab/VLM-R1
CrowdVLM-R1-BiggerDSPaper: https://huggingface.co/papers/2504.03724
Git repo: https://github.com/yeyimilk/CrowdVLM-R1
Citation
@misc{wang2025crowdvlmr1expandingr1ability,
title={CrowdVLM-R1: Expanding R1 Ability to Vision Language Model for Crowd Counting using Fuzzy Group Relative Policy Reward},
author={Zhiqiang Wang and Pengbin Feng and Yanbin Lin and Shuzhang Cai and Zongao Bian and Jinghua Yan and Xingquan Zhu},
year={2025},
eprint={2504.03724},
archivePrefix={arXiv}… See the full description on the dataset page: https://huggingface.co/datasets/yeyimilk/CrowdVLM-R1-BiggerDS.Vision-R1-rlNav-R1Vision-R1-DataThis repository contains the dataset used in the paper Vision-R1: Evolving Human-Free Alignment in Large Vision-Language Models via Vision-Guided Reinforcement Learning.
Code: https://github.com/jefferyZhan/Griffon
qrcode_image_100k_r16_textdrop0.5Remember-R1
[ACM MM 2026] Remember-R1: Mitigating Long-Context Visual Forgetting through Reinforcement Learning
This repository provides the training dataset for Remember-R1.
🔗 Quick Links
📄 Paper
💻 GitHub
🤗 Remember-R1-3B
🤗 Remember-R1-7B
🧠 Remember-R1 Dataset
📌 Citation
If you find Remember-R1 useful in your research, please cite our paper:
@inproceedings{chen2026rememberr1,
title={Remember-R1: Mitigating Long-Context Visual Forgetting through… See the full description on the dataset page: https://huggingface.co/datasets/JM-Chen/Remember-R1.mcot_r1_mcq_66ksole-r1-oxe-annot-1026im2vec-svg-emoji
im2vec-svg-emoji
Rendered (SVG, PNG) pairs for training Im2Vec,
a raster-logo-to-SVG model. This is a pre-rendered version of
starvector/svg-emoji:
each .svg file is paired with a same-stem .png rendered at 256x256 via
cairosvg, matching this project's training pipeline
(im2vec/data/render.py).
Why this dataset
The project's original training data, FIGR-8,
turned out to contain zero color information — every sampled training
SVG relies on the SVG default fill… See the full description on the dataset page: https://huggingface.co/datasets/R1l3y-w/im2vec-svg-emoji.R1-ShareVL-SynGEOQA_8K_R1VClevr_CoGenT_TrainA_R1UI-R1-3B-TrainThis repository contains the training data presented in UI-R1: Enhancing Action Prediction of GUI Agents by Reinforcement Learning.
Project page: https://github.com/lll6gg/UI-R1
Koni_R15_Anime_Girl_Images_ZH_Captioned
synthux-economy-r10-w184
SynthUX Computer-Use Dataset — economy sim r10
Grounded visual computer-use trajectories generated by SynthUX. Each record is
one worker's device session inside a simulated company: a goal expands into a
node tree, terminal nodes drive real desktop-simulator apps (Terminal, Notes,
VS Code, Browser, Slack/Teams, Mail, Sheets, Slides, …) through low-level
mouse/keyboard input, and the observed trajectory is recorded as a screen video
plus per-node frames.
App-native… See the full description on the dataset page: https://huggingface.co/datasets/jacob-valdez/synthux-economy-r10-w184.SK-IBAN-synthetic-r1
Slovak IBAN OCR Synthetic Dataset
Hard samples — obrázky, ktoré Qwen3-VL-4B neprečítal správne,
ale GLM-OCR (zai-org/GLM-OCR) ich prečítal korektne (čitateľné).
Pipeline filtrovania
Generovanie syntetických IBAN obrázkov s degradáciami
Qwen3-VL-4B filter — ponechané iba vzorky, ktoré Qwen neprečítal správne (hard)
GLM-OCR filter — z hard vzoriek ponechané iba tie, ktoré GLM-OCR prečítal správne
(overenie čitateľnosti — nečitateľné obrázky sú zahodené)
Štatistiky… See the full description on the dataset page: https://huggingface.co/datasets/adamgavora/SK-IBAN-synthetic-r1.CrowdVLM-R1-dataPaper: https://huggingface.co/papers/2504.03724
Git repo: https://github.com/yeyimilk/CrowdVLM-R1
Citation
@misc{wang2025crowdvlmr1expandingr1ability,
title={CrowdVLM-R1: Expanding R1 Ability to Vision Language Model for Crowd Counting using Fuzzy Group Relative Policy Reward},
author={Zhiqiang Wang and Pengbin Feng and Yanbin Lin and Shuzhang Cai and Zongao Bian and Jinghua Yan and Xingquan Zhu},
year={2025},
eprint={2504.03724},
archivePrefix={arXiv}… See the full description on the dataset page: https://huggingface.co/datasets/yeyimilk/CrowdVLM-R1-data.