Overcoming Scene Context Constraints for Object Detection in wild using Defilters

TL;DR

Object detection in real-world imagery is hampered by distortions from uncontrolled capture. The authors introduce a distortion-removal pre-processing pipeline that first classifies distortions using $GLCM$ and $LBP$ features with ensemble classifiers and then applies defilters to produce cleaner images before detection, reducing reliance on distorted-training data. Evaluations on a CD-COCO-inspired dataset show that defiltered inputs processed by state-of-the-art detectors (notably InternImage-XL) reach $mAP$ values of $0.562$ on validation and $0.564$ on test, signaling strong performance gains. This approach offers practical impact for real-world vision systems by boosting accuracy without retraining detectors on distorted data, though it notes limitations for local-level distortions and suggests future work to address them. $mAP$ gains demonstrate the viability of distortion-removal as a pre-processing step in robust object detection.

Abstract

This paper focuses on improving object detection performance by addressing the issue of image distortions, commonly encountered in uncontrolled acquisition environments. High-level computer vision tasks such as object detection, recognition, and segmentation are particularly sensitive to image distortion. To address this issue, we propose a novel approach employing an image defilter to rectify image distortion prior to object detection. This method enhances object detection accuracy, as models perform optimally when trained on non-distorted images. Our experiments demonstrate that utilizing defiltered images significantly improves mean average precision compared to training object detection models on distorted images. Consequently, our proposed method offers considerable benefits for real-world applications plagued by image distortion. To our knowledge, the contribution lies in employing distortion-removal paradigm for object detection on images captured in natural settings. We achieved an improvement of 0.562 and 0.564 of mean Average precision on validation and test data.

Figures (2)

• Figure 1: Proposed framework pipeline for object detection
• Figure 2: (a) Object detection on distorted images, (b) Performance improvement achieved by our proposed method: object detection after defiltering images