[Re] Unsupervised Object Segmentation by Redrawing [Rev2], NeurIPS 2019 Reproducibility Challenge

Nabakumar Singh Khongbantabam, Joel Fredriksson

Abstract: This report is part of the NeurIPS 2019 Reproducibility Challenge and attempts to replicate the results of the paper Unsupervised Object Segmentation by Redrawing. We attempt to replicate the results by implementing the deep neural network architecture, named ReDo by the authors in the original paper, based on the specifications described in the paper. ReDo architecture is designed based on a Generative Adversarial Network (GAN) to learn and segment object masks using unsupervised learning. [Rev 2]

Track: Replicability

NeurIPS Paper Id: https://openreview.net/forum?id=r1e95rSlUB