4 papers
TAGRPO: Boosting GRPO on Image-to-Video Generation with Direct Trajectory Alignment
Jin Wang, Jianxiang Lu, Guangzheng Xu +10
Recent studies have demonstrated the efficacy of integrating Group Relative Policy Optimization (GRPO) into flow matching models, particularly for text-to-image and text-to-video g…
VisualNeedle: Benchmarking Active Visual Search in Information-Dense Scenes
Jingru Chen, Yiming Liu, Mingtao Chen +5
Frontier multimodal large language models (MLLMs) have been reported to achieve over 90% accuracy on fine-grained perception benchmarks. However, such scores do not necessarily imp…
Q-Save: Towards Scoring and Attribution for Generated Video Evaluation
Xiele Wu, Zicheng Zhang, Mingtao Chen +7
Evaluating AI-generated video (AIGV) quality hinges on three crucial dimensions: visual quality, dynamic quality, and text-video alignment. While numerous evaluation datasets and a…
Rethinking KL Regularization in RLHF: From Value Estimation to Gradient Optimization
Kezhao Liu, Jason Klein Liu, Mingtao Chen +1
Reinforcement Learning from Human Feedback (RLHF) leverages a Kullback-Leibler (KL) divergence loss to stabilize training and prevent overfitting. However, in methods such as GRPO,…