We report quantitative results by four numeric metrics, i.e., PSNR [33], SSIM [5], flow warping error [17] and video-based Fr´echet Inception Distance (VFID) [5,30]. Specifically, we use PSNR and SSIM as they are the most widely-used metrics for video quality assessment. Besides, the flow warping error is included to measure the temporal stability of generated videos. Moreover, FID has been proved to be an effective perceptual metric and it has been used by many inpainting models [25,30,38]. In practice, we use an I3D [4] pre-trained video recognition model to calculate VFID following the settings in [5,30].
PSNR 和 SSIM ：
使用方法：
图像质量评价指标： PSNR 和 SSIM_马鹏森的博客-CSDN博客_ssim和psnr哪个好
Flow warping error :
Temporal stability ： We measure the temporal stability of a video based on the flow warping error between two frames:（https://arxiv.org/pdf/1808.00449v1.pdf）
使用方法：（GitHub - phoenix104104/fast_blind_video_consistency: Learning Blind Video Temporal Consistency (ECCV 2018)）
video-based Fr´echet Inception Distance (VFID)：
FID介绍：FID使用（Frechet Inception Distance score）_马鹏森的博客-CSDN博客
VFID使用方法：
Free-Form-Video-Inpainting/evaluate.py at master · amjltc295/Free-Form-Video-Inpainting · GitHub
在evalute.py里面有评价指标的代码
17. Lai, W.S., Huang, J.B., Wang, O., Shechtman, E., Yumer, E., Yang, M.H.: Learning blind video temporal consistency. In: ECCV. pp. 170–185 (2018) 5. Chang, Y.L., Liu, Z.Y., Lee, K.Y., Hsu, W.: Free-form video inpainting with 3d gated convolution and temporal patchgan. In: ICCV. pp. 9066–9075 (2019) 30. Wang, T.C., Liu, M.Y., Zhu, J.Y., Liu, G., Tao, A., Kautz, J., Catanzaro, B.: Video-to-video synthesis. In: NeuraIPS. pp. 1152–1164 (2018)
