A Fair Ranking and New Model for Panoptic Scene Graph Generation

TL;DR

This paper corrects evaluation flaws in panoptic scene graph generation, introduces a new two-stage model called DSFormer that outperforms existing models, and emphasizes the importance of proper evaluation protocols.

Contribution

It provides a fair evaluation protocol for PSGG, demonstrates the competitiveness of two-stage methods, and introduces the DSFormer model that achieves state-of-the-art results.

Findings

Existing evaluation protocols can be exploited to inflate scores.

Two-stage methods are competitive with one-stage methods when evaluated fairly.

DSFormer significantly outperforms previous models on the corrected benchmark.

Abstract

In panoptic scene graph generation (PSGG), models retrieve interactions between objects in an image which are grounded by panoptic segmentation masks. Previous evaluations on panoptic scene graphs have been subject to an erroneous evaluation protocol where multiple masks for the same object can lead to multiple relation distributions per mask-mask pair. This can be exploited to increase the final score. We correct this flaw and provide a fair ranking over a wide range of existing PSGG models. The observed scores for existing methods increase by up to 7.4 mR@50 for all two-stage methods, while dropping by up to 19.3 mR@50 for all one-stage methods, highlighting the importance of a correct evaluation. Contrary to recent publications, we show that existing two-stage methods are competitive to one-stage methods. Building on this, we introduce the Decoupled SceneFormer (DSFormer), a novel two-stage model that outperforms all existing scene graph models by a large margin of +11 mR@50 and +10 mNgR@50 on the corrected evaluation, thus setting a new SOTA. As a core design principle, DSFormer encodes subject and object masks directly into feature space.