Knowing the Unknown: Interpretable Open-World Object Detection via Concept Decomposition Model

TL;DR

This paper proposes a concept-driven InterPretable OWOD framework by introducing a Concept Decomposition Model (CDM) for OWOD, which explicitly decomposes the coupled RoI features in Faster R-CNN into discriminative, shared, and background concepts.

Abstract

Open-world object detection (OWOD) requires incrementally detecting known categories while reliably identifying unknown objects. Existing methods primarily focus on improving unknown recall, yet overlook interpretability, often leading to known-unknown confusion and reduced prediction reliability. This paper aims to make the entire OWOD framework interpretable, enabling the detector to truly "knowing the unknown". To this end, we propose a concept-driven InterPretable OWOD framework(IPOW) by introducing a Concept Decomposition Model (CDM) for OWOD, which explicitly decomposes the coupled RoI features in Faster R-CNN into discriminative, shared, and background concepts. Discriminative concepts identify the most discriminative features to enlarge the distances between known categories, while shared and background concepts, due to their strong generalization ability, can be readily transferred to detect unknown categories. Leveraging the interpretable framework, we identify that known-unknown confusion arises when unknown objects fall into the discriminative space of known classes. To address this, we propose Concept-Guided Rectification (CGR) to further resolve such confusion. Extensive experiments show that IPOW significantly improves unknown recall while mitigating confusion, and provides concept-level interpretability for both known and unknown predictions.

Figures (6)

• Figure 1: Left: The proposed IPOW reformulates OWOD as a problem of concept decomposition modeling, consisting of discriminative concepts responsible for known-class recognition, along with shared and background concepts that support unknown object detection. Right: Within the IPOW framework, known–unknown confusion can be viewed as unknown objects falling into the discriminative space learned for known classes.
• Figure 2: Overview of the IPOW framework. GMM-RPN is utilized to mitigate the proposal generation bias towards known-categories. A concept head is adopted before decomposition. Each RoI feature is decomposed into discriminative, shared, and background concepts for known object recognition, transferable unknown discovery, and contextual modeling.
• Figure 3: Score distributions on M-OWODB Task 1 for known, confusion (unknown falsely predicted as known), and unknown samples. Discriminative concept activations (left) and the geometric mean of class-specific shared concept activations (right).
• Figure 4: Ablation study on the number of shared concepts on M-OWODB. $K$ and $M$ denote the numbers of LLM-derived shared concepts and residual shared concepts, respectively.
• Figure 5: Qualitative results contrasting IPOW with RandBox, and CROWD, demonstrating that IPOW offers interpretable concept-level reasoning, reduces known--unknown confusion, and enables effective generalization to unknown objects.