DeCo-DETR: Decoupled Cognition DETR for efficient Open-Vocabulary Object Detection

2026-04-03

AI Summary

This work addresses the high computational cost of existing open-vocabulary object detection (OVOD) methods, which rely on online text encoders during inference, and the inherent trade-off between closed-set accuracy and open-world generalization due to entangled training objectives. To overcome these limitations, the authors propose DeCo-DETR, a novel decoupled cognition paradigm that leverages a pretrained large vision-language model (LVLM) to generate region descriptions, which are then aligned via CLIP to construct a hierarchical semantic prototype space. By disentangling semantic representation learning from localization, DeCo-DETR eliminates the need for online text encoding, substantially improving inference efficiency. The method achieves competitive zero-shot detection performance on standard OVOD benchmarks while simultaneously maintaining strong closed-set detection accuracy and robust open-vocabulary generalization.

Abstract

Open-vocabulary Object Detection (OVOD) enables models to recognize objects beyond predefined categories, but existing approaches remain limited in practical deployment. On the one hand, multimodal designs often incur substantial computational overhead due to their reliance on text encoders at inference time. On the other hand, tightly coupled training objectives introduce a trade-off between closed-set detection accuracy and open-world generalization. Thus, we propose Decoupled Cognition DETR (DeCo-DETR), a vision-centric framework that addresses these challenges through a unified decoupling paradigm. Instead of depending on online text encoding, DeCo-DETR constructs a hierarchical semantic prototype space from region-level descriptions generated by pre-trained LVLMs and aligned via CLIP, enabling efficient and reusable semantic representation. Building upon this representation, the framework further disentangles semantic reasoning from localization through a decoupled training strategy, which separates alignment and detection into parallel optimization streams. Extensive experiments on standard OVOD benchmarks demonstrate that DeCo-DETR achieves competitive zero-shot detection performance while significantly improving inference efficiency. These results highlight the effectiveness of decoupling semantic cognition from detection, offering a practical direction for scalable OVOD systems.

Problem

• Open-Vocabulary Object Detection
• computational overhead
• tightly coupled training
• inference efficiency
• generalization

Innovation

• Decoupled Cognition
• Open-Vocabulary Object Detection
• Semantic Prototype Space
• Efficient Inference
• Vision-Centric Framework