ProCal: Inference-Time Proposal Calibration for Open-Vocabulary Object Detection

2026-07-02

AI Summary

This work addresses the limitation in open-vocabulary object detection where classification scores from vision-language models (VLMs) lack localization awareness, leading to inaccurate ranking of proposals for unseen categories. To resolve this, the authors propose ProCal, a proposal calibration method that operates at inference time without requiring fine-tuning. ProCal leverages, for the first time, the foreground–background discrimination capability of frozen VLMs—such as CLIP ViT-L/14—to combine a localization-aware foreground score with a background suppression score, thereby re-ranking detection proposals. Evaluated on OV-LVIS, ProCal improves APr by 2.5 points, substantially suppresses background false activations, and elevates the ranking priority of genuine proposals corresponding to novel categories.

Abstract

Open-vocabulary object detection aims to localize and classify objects beyond the fixed set of categories seen dur ing training. Recent open-vocabulary object detection methods improve localization and classification for unseen categories by leveraging a frozen VLM as a detector backbone. However, VLM classification score lacks recognizing position and scale of the object in an image. We observe that pretrained VLMs en able to classify foreground and background regions. According to this observation, we propose a simple inference-time Pro posal Calibration (ProCal) that improves localization quality of the classification score. ProCal computes a proposal prior by combining two scores: localization-aware foreground score and background-aware suppression score. Localization-aware foreground score captures whether a proposal contains an object area. Background-aware suppression score measures the extent to which the proposal resembles background. We analyze that ProCal suppresses false novel activation on background proposals and consistently ranks true novel proposals above background and partial novel proposals. Applied to CLIPSelf ViT-L/14, ProCal improves APr +2.5 on OV-LVIS. The analyses show that proposal-level localization-aware reranking effects to mitigate ranking miscalibration for novel objects.

Problem

• open-vocabulary object detection
• visual-language model
• localization
• proposal calibration
• novel object detection

Innovation

• Proposal Calibration
• Open-Vocabulary Object Detection
• Vision-Language Model
• Inference-Time Calibration
• Localization-Aware Reranking