Reproduce “Smart-Insertion-V: Photorealistic Video Insertion via a Closed-Loop Feedback Dual-Stream Framework” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xiao Cao, Yansong Qu, Xiangzhen Chang, Wen Xiao, Jiakui Hu, Heyuan Li, Jialun Liu, Zhiyong Huang and 1 more
Smart-Insertion-V integrates reference appearance adaptation with video object insertion through parallel image-style-transfer and video-generation streams, whose intermediate features guide video generation. At inference, a vision-language model uses an intermediate image estimate to refine guidance. Its ChatGPT-evaluated Harmonic Score is 0.842, versus 0.693 for UniVideo-Q; a single-stream ablation scores 0.474. These results are from the reported setup, with limited spatial control.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.