MrNeRF (@janusch_patas)

2024-11-23 | โค๏ธ 167 | ๐Ÿ” 22


ViewExtrapolator: Novel View Extrapolation with Video Diffusion Priors

Code is already released! Also make sure to checkout the original authorโ€™s post!

Contributions (cited):

  • We introduce ViewExtrapolator, a novel training-free pipeline that leverages the generative priors of SVD for novel view extrapolation.

  • We design guidance annealing and resampling annealing that eliminate artifacts and enable high-quality inpainting of unseen regions, enhancing the visual fidelity of the rendered novel views effectively.

  • Extensive experiments over various 3D rendering approaches demonstrate the superiority and broad applicability of ViewExtrapolator in novel view extrapolation

๋ฏธ๋””์–ด

video

์ธ์šฉ

[@SongNeverPanic]: 1/ ๐Ÿš€ Excited to share our new work: Novel View Extrapolation with Video Diffusion Priors. Dive into how we push the boundaries of novel view synthesis with ViewExtrapolator, a training-free meโ€ฆ


์ธ์šฉ ํŠธ์œ—

Kunhao (@SongNeverPanic)

1/ ๐Ÿš€ Excited to share our new work: Novel View Extrapolation with Video Diffusion Priors. Dive into how we push the boundaries of novel view synthesis with ViewExtrapolator, a training-free method that leverages Stable Video Diffusion (SVD). ๐Ÿงต ๐Ÿ”—https://t.co/ObLFAVYSot https://t.co/oPCKwgwPWQ

์›๋ณธ ํŠธ์œ—

๐ŸŽฌ ์˜์ƒ ๐ŸŽฌ ์˜์ƒ ๐ŸŽฌ ์˜์ƒ

Tags

domain-rendering domain-genai domain-ai-ml