Preprint: https://www.biorxiv.org/content/early/2018/04/20/304774
Reprint: https://ieeexplore.ieee.org/document/8616183
The thumbnail images below show the pixel-wise average over all frames of the associated video.
This shows a composite video of the trained generator samples conditioned on the surrogate "ext" images (See Fig. 1 for a description of the datasets). This demonstrate the limit of the reconstruction accuracy under our method.
This shows a composite video of the trained generator samples conditioned on the denoised vim-1 "val" voxel set corresponding to Fig. 4. This demonstrate the generality of the decoder and the limit of reconstruction accuracy given successful denoising.
This shows a composite video of the trained generator samples conditioned on the fully cross validated vim-1 "test" voxel set corresponding to Fig. 5.
Learning the feature extractor