140f6969d5213fd0ece03148e62e461e-AuthorFeedback.pdf

Model J (Mean) F (Mean) FlowNet2[16] 26.7 25.2 PWC-Net[39] 35.2 37.4 Ours 57.7 60.0 Table3: Ablationstudyontemperature(notrackintesting). Temperature J (Mean) F (Mean) 1 56.8 59.5 2 52.3 56.2 10 51.0 55.5 Which methods should the work compare with? Rather, it aims to achieve the same goal as presented in [43] and [49] - learning unsupervised27 correspondences. These two tasks are significantly different: one focuses on learning a feature matching network28 that can track regions and pixels between frames over a long period, while the other emphasizes modeling subtle29 displacements betweenadjacentframes.