NeurIPS 2020

Exploiting weakly supervised visual patterns to learn from partial annotations

Meta Review

This paper received reviews from 4 reviewers. The reviewers appreciated the solid, clear exploration of partial annotations, an important problem. The proposed methods are sensible and have positive simplicity. The reviewers engaged in further discussion after reading the authors' rebuttal. Overall, while additional experimentation (e.g. against GCNs) is possible, the current manuscript already provides a substantial amount of empirical work. The demonstration of the effectiveness of these simple methods outperforming more advanced baselines is valuable to the community, and the paper should spur continued interest in this important partial annotation problem space.