Reward models, preference optimization, AI-image detection and video evaluation: researchers train and test their models on human judgments collected with Rapidata, from our open datasets on Hugging Face or from studies run directly on the platform.
ACL
With authors from NVIDIA, Google Cloud AI, Amazon, Adobe, ByteDance, Tencent Hunyuan, Alibaba, Hugging Face, Kuaishou Kling, Ant Group, Krea AI, Columbia University, KAIST, University of Waterloo, CUHK, UIUC, UT Austin and 26 more institutions.
Papers built on Rapidata human feedback
Accepted at peer-reviewed venues
Researchers across the author lists
Rapidata open datasets in use
Every entry was checked against the paper's full text.
Tencent Hunyuan · Fudan University · Shanghai AI Lab
“OpenAI-4o_t2i_human_preference (6.7K) collected by Rapidata… Text2Video-Human Preferences (5.7K), also collected by Rapidata”
NVIDIA · University of Waterloo · Vector Institute
“Rapidata-I2V is a pairwise preference dataset released on HuggingFace datasets by Rapidata”
CUHK MMLab · Hugging Face · ByteDance +3
“We leverage Rapidata to collect human preference annotations.”
Chinese Academy of Sciences
Benchmark · Benchmark construction across 27 text-to-image models
Shanghai AI Lab · CUHK · Krea AI +1
Leaderboard · Reports Elo rankings from the Rapidata text-to-image leaderboard
ByteDance · Northeastern University
Training · Generative reward-model training
Shanghai AI Lab · CUHK MMLab · Krea AI +2
Human study · Pairwise human preference study on the Rapidata platform
Shanghai AI Lab · Fudan University
Training · Reward-model training
Adobe · IIIT-Delhi · SUNY Buffalo
Evaluation · Personalization benchmark with annotator demographics
Shanghai Jiao Tong University · Ant Group
Evaluation · Out-of-distribution evaluation
Kuaishou Kling · CUHK MMLab · Nanjing University
Training · Video reward-model training
University of Waterloo · HKUST · Alibaba
Training · Reward-model training
Fudan University · University of Adelaide
Training · Preference optimization of Stable Diffusion 1.5
UT Austin
Human study · Human evaluation run on Rapidata
Arizona State University
Training · Diffusion-DPO on Stable Diffusion 2.1
KAIST · Google Cloud AI
Benchmark · Benchmark construction
Amazon · Georgia Tech
Evaluation · Evaluation on Rapidata human judgments of generated images
Shanghai AI Lab · Hugging Face · Krea AI +2
Human study · Video study with a screened Rapidata audience
UIUC · Columbia University · MBZUAI
Human study · Data annotation on Rapidata
Evaluation · Eye-tracking study on Rapidata-labelled image pairs
Fudan University · Shanghai AI Lab
Training · Personalized reward-model training
Tencent Youtu Lab · Shanghai Jiao Tong University
Training · Text-to-video reward-model training
Shanghai AI Lab · Shanghai Jiao Tong University
Human study · Human study on the Rapidata platform
Shanghai Jiao Tong University · Ant Group
Evaluation · Generalization test set
Training · Multimodal judge training (12K pairs)
Analysis · Analysis of 40,000 AI-generated images
Training · Video reward-model training
Universidad Autónoma de Madrid
Evaluation · Evaluation images
MWS AI · ITMO University
Evaluation · Evaluation prompts
Papers that reference our research and writing on human evaluation.
Cites Rapidata's 2M-vote study on why automatic metrics miss human preference
Cites Rapidata's work on rich human feedback for text-to-image models
Cites Rapidata's 2M-vote study on why automatic metrics miss human preference
Cites Rapidata's 2M-vote study on human evaluation of generative models
Cites Rapidata's 2M-vote model comparison
Cites a Rapidata blog post on fine-tuning Stable Diffusion
Send us the link. Need human feedback for your next study? Our research grants cover up to $50,000 in annotation credits.