RenDetNet: Weakly-supervised Shadow Detection with Shadow Caster Verification

TL;DR

RenDetNet introduces a self-supervised shadow detection method that verifies detected shadows by physically re-rendering scenes to ensure they have corresponding shadow casters, improving accuracy without extensive annotations.

Contribution

It is the first learning-based shadow detection model with self-supervised supervisory signals derived from differentiable scene re-rendering.

Findings

Outperforms recent models on our dataset

First self-supervised shadow detection approach

Code released on GitHub

Abstract

Existing shadow detection models struggle to differentiate dark image areas from shadows. In this paper, we tackle this issue by verifying that all detected shadows are real, i.e. they have paired shadow casters. We perform this step in a physically-accurate manner by differentiably re-rendering the scene and observing the changes stemming from carving out estimated shadow casters. Thanks to this approach, the RenDetNet proposed in this paper is the first learning-based shadow detection model whose supervisory signals can be computed in a self-supervised manner. The developed system compares favourably against recent models trained on our data. As part of this publication, we release our code on github.