Anticipating Future Object Compositions without Forgetting

TL;DR

This paper uses Grounding DINO and incorporates Compositional Soft Prompting into Compositional Soft Prompting into it and extends it with Compositional Anticipation to enhance CZSL in object detection without forgetting prior learned knowledge.

Abstract

Despite the significant advancements in computer vision models, their ability to generalize to novel object-attribute compositions remains limited. Existing methods for Compositional Zero-Shot Learning (CZSL) mainly focus on image classification. This paper aims to enhance CZSL in object detection without forgetting prior learned knowledge. We use Grounding DINO and incorporate Compositional Soft Prompting (CSP) into it and extend it with Compositional Anticipation. We achieve a 70.5% improvement over CSP on the harmonic mean (HM) between seen and unseen compositions on the CLEVR dataset. Furthermore, we introduce Contrastive Prompt Tuning to incrementally address model confusion between similar compositions. We demonstrate the effectiveness of this method and achieve an increase of 14.5% in HM across the pretrain, increment, and unseen sets. Collectively, these methods provide a framework for learning various compositions with limited data, as well as improving the performance of underperforming compositions when additional data becomes available.

Figures (2)

• Figure 1: Our method anticipates unseen, future object-attribute compositions through Compositional Independence and Compositional Smoothing. Forgetting is mitigated by creating auxiliary tokens for the language embeddings and refining only these tokens. Errors in compositions are incrementally corrected using Contrastive Prompt Tuning, which contrasts confused compositions.
• Figure 2: Our Contrastive Prompt Tuning method is effective in incremental learning. It improves performance on the increment compositions (in gray) while preserving performance of the pretrained compositions (in black).