Dynamic Scene Reconstruction and Compression

The field of computer vision and graphics is moving towards more efficient and accurate methods for dynamic scene reconstruction and compression. Recent developments have focused on novel view synthesis, motion-aware neural rendering, and sparse multi-view dynamic Gaussian Splatting. These approaches have shown promising results in reconstructing dynamic scenes from monocular videos and improving the quality of reconstructed images. Noteworthy papers include: 4D3R, which introduces a pose-free dynamic neural rendering framework that achieves up to 1.8dB PSNR improvement over state-of-the-art methods. Splatography, which presents an approach that splits the canonical Gaussians and deformation field into foreground and background components, producing state-of-the-art qualitative and quantitative results. Modulo Video Recovery via Selective Spatiotemporal Vision Transformer, which demonstrates the first deep learning framework for modulo video reconstruction, achieving state-of-the-art performance in modulo video recovery. Burst Image Quality Assessment, which proposes a new task of Burst Image Quality Assessment to evaluate the task-driven quality of each frame within a burst sequence. Machines Serve Human, which sets out a novel collaborative compression method based on the machine-vision-oriented compression. Neural B-frame Video Compression with Bi-directional Reference Harmonization, which proposes a novel NBVC method with the proposed Bi-directional Motion Converge and Bi-directional Contextual Fusion.

Sources

4D3R: Motion-Aware Neural Reconstruction and Rendering of Dynamic Scenes from Monocular Videos

Splatography: Sparse multi-view dynamic Gaussian Splatting for filmmaking challenges

Modulo Video Recovery via Selective Spatiotemporal Vision Transformer

Burst Image Quality Assessment: A New Benchmark and Unified Framework for Multiple Downstream Tasks

Machines Serve Human: A Novel Variable Human-machine Collaborative Compression Framework

Neural B-frame Video Compression with Bi-directional Reference Harmonization

Built with on top of