目录
DVC: An End-to-end Deep Video Compression Framework
train on Vimeo-90kVideo enhancement with task-oriented flow
evaluate on UVGUltra video group test sequences and HEVC Standard Test Sequences(Class B,C,D,E)Overview of the high efficiency video coding(hevc) standard
PSNR and MS-SSIM
training image (256 imes 256)
2 Titan X GPUs 7 days
Learned Video Compression with Feature-level Residuals
pre-train on Vimeo-90k
fine-tune on CLIC dataset
2 GTX 1080ti GPUs 7 days
End-to-end Optimized Video Compression with MV-Residual Prediction
evaluation on CLIC 2020 P-frame validation datasets
Scale-space flow for end-to-end optimized video compression
模型接受了视频帧的训练，这些视频是从大约700000个高清(1920×1080)的视频中提取出来的，帧速率为30Hz(YouTube已经对这些视频进行了编码)
Video Compression through Image Interpolation
train our models using videos from the Kinetics dataset [7], train on 27k, 5k for validation, 5k for testing
additionally test on VTL and UVG