BiFormer: Learning Bilateral Motion Estimation via Bilateral Transformer for 4K Video Frame Interpolation

Junheum Park; Jintae Kim; Chang-Su Kim

doi:10.1109/cvpr52729.2023.00157

CVPR2023顶会

BiFormer: Learning Bilateral Motion Estimation via Bilateral Transformer for 4K Video Frame Interpolation

Junheum Park, Jintae Kim, Chang-Su Kim

arXiv DOI 出版方

2023年份

16顶会引用

摘要

233px 163px 𝒱 𝑡→1 𝒱 𝑡→1 𝒱 𝑡→1 23.67dB / 0.8137 26.64dB / 0.8634 20.21dB / 0.7516 Blended Input Figure 1. Examples of 4K video frame interpolation results, obtained by ABME [1], XVFI [2], and the proposed BiFormer. 4K video frame interpolation is challenging due to large motion magnitudes, e.g. hundreds of pixels. PSNR/SSIM scores are presented within the interpolation results, and the estimated motion fields Vt→1 are at the bottom row.