Reproduce “ImmersePro: End-to-End Stereo Video Synthesis Via Implicit Disparity Learning” 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
Jian Shi, Zhenyu Li, Peter Wonka
Researchers from King Abdullah University of Science and Technology (KAUST) developed ImmersePro, an end-to-end framework for synthesizing stereo video from monocular input, and introduced the YouTube-SBS dataset. ImmersePro employs implicit and layered disparity learning to generate temporally consistent right-view frames, achieving superior visual quality and quantitative performance against state-of-the-art methods.
Key insights
- Implicit disparity guidance is critical for the convergence of the model, enabling the learning of stereo effects without direct, potentially error-prone, explicit disparity maps.
- A layered disparity representation, coupled with a specific pixel selection algorithm, effectively mitigates blurring artifacts and improves clarity in the generated stereo views, outperforming direct implicit disparity usage.
- Integrating spatial-temporal attention and a context fusion module across different feature branches is essential for maintaining temporal consistency and enriching semantic details in the synthesized stereo videos.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.