Depth estimation from single-shot monocular endoscope image using image domain adaptation and edge-aware depth estimation

We propose a depth estimation method from single-shot monocular endoscopic image using Lambertian surface translation by domain adaptation and multi-scale edge loss. employ two-step process including unpaired data estimation. The texture specular reflection on the of an organ reduce accuracy estimations. apply to remove these reflections. Then, we estimate fully convolutional network (FCN). During training FCN, improvement object similarity between estimated ground truth is important for getting better results. introduced muti-scale loss function improve quantitatively evaluated proposed real colonoscopic images. values were proportional values. Furthermore, applied images automated anatomical location identification neural network. improved 69.2% 74.1%