This paper introduces a method for synthesizing colorized novel views from grey-scale multi-view images using neural radiance field (NeRF) based methods. The researchers found that applying colorization methods to the generated grey-scale novel views resulted in artifacts due to inconsistency across views. Training a radiance field network on the colorized grey-scale image sequence did not solve the 3D consistency issue. To address this, the authors propose a distillation based method that transfers color knowledge from existing 2D colorization methods to the radiance field network. The experimental results show that their method produces superior colorized novel views with cross-view consistency compared to baseline methods. The method is also effective for colorizing radiance field networks trained from infra-red multi-view images and old grey-scale multi-view image sequences.
https://arxiv.org/abs/2309.07668