Instance Shadow Detection

TL;DR

This paper introduces the novel problem of instance shadow detection, presents a new dataset SOBA, and proposes an end-to-end framework LISA for automatic shadow-object pairing and light direction estimation.

Contribution

The paper provides the first dataset for instance shadow detection and develops LISA, a framework that jointly predicts shadows, objects, associations, and light direction.

Findings

LISA achieves high accuracy on the SOBA dataset.

The new shadow-object average precision metric effectively evaluates performance.

Method demonstrates applicability in light estimation and photo editing.

Abstract

Instance shadow detection is a brand new problem, aiming to find shadow instances paired with object instances. To approach it, we first prepare a new dataset called SOBA, named after Shadow-OBject Association, with 3,623 pairs of shadow and object instances in 1,000 photos, each with individual labeled masks. Second, we design LISA, named after Light-guided Instance Shadow-object Association, an end-to-end framework to automatically predict the shadow and object instances, together with the shadow-object associations and light direction. Then, we pair up the predicted shadow and object instances, and match them with the predicted shadow-object associations to generate the final results. In our evaluations, we formulate a new metric named the shadow-object average precision to measure the performance of our results. Further, we conducted various experiments and demonstrate our method's applicability on light direction estimation and photo editing.