Bridge Past and Future: Overcoming Information Asymmetry in Incremental Object Detection

TL;DR

This paper introduces 'Bridge Past and Future' (BPF), a novel method for incremental object detection that aligns models across stages and leverages background probability to reduce forgetting and improve performance on benchmarks.

Contribution

The paper proposes BPF, a new approach that aligns models across stages and introduces DwF loss to better handle object co-occurrence and mitigate forgetting in incremental detection.

Findings

BPF outperforms state-of-the-art methods on Pascal VOC and MS COCO.

The DwF loss effectively leverages background probability to improve learning.

The method works well without memory in incremental detection scenarios.

Abstract

In incremental object detection, knowledge distillation has been proven to be an effective way to alleviate catastrophic forgetting. However, previous works focused on preserving the knowledge of old models, ignoring that images could simultaneously contain categories from past, present, and future stages. The co-occurrence of objects makes the optimization objectives inconsistent across different stages since the definition for foreground objects differs across various stages, which limits the model's performance greatly. To overcome this problem, we propose a method called ``Bridge Past and Future'' (BPF), which aligns models across stages, ensuring consistent optimization directions. In addition, we propose a novel Distillation with Future (DwF) loss, fully leveraging the background probability to mitigate the forgetting of old classes while ensuring a high level of adaptability in learning new classes. Extensive experiments are conducted on both Pascal VOC and MS COCO benchmarks. Without memory, BPF outperforms current state-of-the-art methods under various settings. The code is available at https://github.com/iSEE-Laboratory/BPF.