[Paper Review] A Middlebury Benchmark & Context-Aware Synthesis for Video Frame Interpolation
- 1. Optical Flow Based Frame Interpolations 朱語所鍵 覈 2018-2-1st 覦 觜 語 郁規 危覩 & Middlebury Benchmark
- 2. Frame Interpolation 螻覲 AdaConv 2017.6 CVPR Niklaus Adaptive Convolution SepConv 2017.10 ICCV Niklaus Phase Based Phase-Based 2015.6 CVPR Meyer PhaseNet 2018.6 CVPR Meyer Optical Flow Based ETCDeep Voxel Flow 2017.10 ICCV Liu Moving Gradient 2009.8 ACM Graphics Mahajan Middlebury 2011.11 IJCV Baker CtxSyn 2018.6 CVPR Niklaus
- 3. Middlebury Benchmark Optical Flow Evaluation 一讌 朱 Evaluation 覦覯 譴, Optical Flow襯 伎 Frame Interpolation 焔レ 蠍一朱 螳 覦覯 譟伎 Optical Flow襯 螳蠍 覈朱 Optical Flow Based Frame Interpolation 螻襴讀 覿磯朱 Frame Interpolation 朱語朱 螳覦螻 . Baker, Simon, et al. "A database and evaluation methodology for optical flow." International Journal of Computer Vision 92.1 (2011): 1-31.
- 4. Middlebury Benchmark 企語 0, 1 伎 Optical Flow 0 譟伎 Optical Flow 0 襯 伎覃 0~1 t Optical Flow °ゼ 詞 . round x + tu0 x = u0(x) 覿覈 ′ 企 讌 u0 螳 伎 覈磯Υ 螻, 企 讌 覓企 覦一讌 . u0(x1)螻 u0(x2)螳 蟆曙 蟆曙, |0 1( + 0())| 螳 讓曙 豈. 觜 螻褐 Outside-in strategy襯 伎.
- 5. Middlebury Benchmark Occlusion Mask 0 , 1()襯 蟲. = 1企朱 詞 腫()曙 覦ｸ 覲伎伎 る 詞企. round x + tu0 x = u0(x) = 1 覃 u1 蟲 , 企 蠍 hole れ 1 = 1襦 . 0 蟆曙, 0 1 + 0 > 0.5企 0 = 1襦 . 伎 腫()曙 螳 蟲蠍 谿語＾ 0 豺 0 1 豺 1襯 蟲伎 . 讀 0(0) 1(1) 谿 譟壱 蟆企. 0 = ＄ () 1 = + 1 () 腫 = 1 0 0 + ＄1(1) 襷 0 0 = 1企朱 腫() = 0 0 覦 蟆曙磯 襷谿螳讌
- 6. Evaluation れ Mask 伎 RMS Error 螻 煙 http://vision.middlebury.edu/flow/eval/results/results-i1.php
- 7. Context Aware Synthesis 腫 = 1 0 0 + ＄1(1) 腫 = 0(0) Forward Wrapped Image 腫 = 1(1) Backward Wrapped Image 企語襯 Linear Combination(Blending)讌 襷螻, Neural Network襯 伎伎 企慨! Pre-trained Network 詞 Feature蟾讌 Wrap 伎 螳 ｌ伎殊!! Niklaus, Simon, and Feng Liu. "Context-aware Synthesis for Video Frame Interpolation." arXiv preprint arXiv:1803.10967(2018).
- 8. Future Plan 伎 Related Works る 螻襴讀れ 覿覿 Flow-Based蟇磯 企 曙 朱碁 覈覈 朱碁 觜襯願 襷覓企Μ螻 讌 蟲 螳 螻襴讀襷 讌 覈讌 殊 企至 蟲 讌 Frame Interpolation 蟯 朱 TCSVT 覈襦 ! 危 Optical Flow襯 蟇企 覲 螳