Any-scale Object Detection using Arbitrary-scaled Images

2025-07-26

2025 19th International Conference on Machine Vision and Applications (MVA)

AI Summary

This study addresses the high false positive rate caused by independently applying super-resolution images in arbitrary-scale object detection. To overcome this limitation, we propose a continuous-scale detection framework that integrates super-resolution with scale proposal maps. Unlike conventional approaches constrained to discrete multi-scale representations, our method predicts scale proposal maps to precisely identify pixel regions suitable for each super-resolution scale, thereby enabling continuous image rescaling and subsequent object detection. By selectively aligning resolution enhancement with region-specific scale requirements, this work effectively suppresses spurious positive detections. Experimental results demonstrate that the proposed framework significantly improves object localization accuracy across diverse cross-scale scenarios, offering a robust solution for arbitrary-scale detection tasks.

Abstract

This paper proposes any-scale object detection using arbitrary-scale super-resolution for continuously rescaling object images, while general multi-scale object detection uses discretely rescaled appearance representations. However, a naïve usage of super-resolution produces many false-positive detections if many super-resolution images are independently fed into an object detector. Our method suppresses these false positives by predicting scale proposal maps, each of which represents a set of pixels appropriate for each super-resolution scale.

Problem

• Any-scale Object Detection
• Arbitrary-scale Super-resolution
• False-positive Detections
• Multi-scale Object Detection

Innovation

• Any-scale Object Detection
• Arbitrary-scale Super-resolution
• Scale Proposal Maps
• False Positive Suppression