Bidirectional Optical Flow NeRF: High Accuracy and High Quality under Fewer Views

Neural Radiance Fields (NeRF) can implicitly represent 3D-consistent RGB images and geometric by optimizing an underlying continuous volumetric scene function using a sparse set of input views, which has greatly benefited view synthesis tasks. However, NeRF fails to estimate correct geometry when given fewer resulting in failure synthesize novel views. Existing works rely on introducing depth or adding estimation networks resolve the problem poor synthetic with due lack spatial consistency single-depth image performance existing methods still have challenges addressing this problem. So paper proposes Bidirectional Optical Flow NeRF(BOF-NeRF), addresses mining optical flow information between 2D images. Our key insight is that utilizing design loss effectively guide learn right view. We also propose view-enhanced fusion method based color solve details NeRF. conduct extensive experiments NeRF-LLFF DTU MVS benchmarks for tasks different complex real scenes. further demonstrate robustness BOF-NeRF under baseline distances Middlebury dataset. In all cases, outperforms current state-of-the-art baselines estimation.