Abstract
Object detectors often suffer from multiple performance limitations which may be attenuated with larger training datasets, improved training techniques, and complex detection models. However, such strategies are complex and time-consuming for applications requiring fast deployments. We propose a Simple Fusion of Object Detectors (SFOD) late ensemble method to combine existing pre-trained, off-the-shelf, fine-tuned object detectors and leverage on their divergences to improve the overall detection performance. Comprehensive experimental evaluations, based on PASCAL VOC07 challenge, demonstrate SFOD’s ability to improve mean average precision ( ${mAP}$ ) for different fusion sizes and base detector combinations, reaching an absolute 84.08% ${mAP}$ and an improvement of 3.97% ${mAP}$ . The improvements extend to most classes, fusion sizes, and base detector combinations, revealing $AP$ improvements up to 17.35% over baselines, for particular object classes. Practical application evaluations, based on optimal threshold selection, also reveal improvements of 10.54% and 8.36% of mean recall ( $mR$ ) and ${mAP}$ , respectively. Our approach does not require additional training and is quickly deployable, yet providing a few adjustable hyperparameters to optimize the recall-precision relation for specific applications. Improvements obtained from our proposed SFOD fusion pipeline span across a broad range of object classes and are important for a wide variety of critical applications where every successful detection is treasured.
Highlights
EXPERIMENTAL RESULTS Experimental evaluations conducted in this work, are focused on three main perspectives: (i) standard PASCAL VOC07 detection performance, (ii) practical applications performance and, (iii) missed detections
EXPERIMENT DESIGN Before performing the experimental evaluation of the proposed Simple Fusion of Object Detectors (SFOD) fusion pipeline, we (i) determine object detectors’ pairwise divergences to assess the likelihood of obtaining detection performance improvements from the proposed SFOD fusion pipeline, (ii) assess the ability of the proposed fusion method to reduce the number of undetected objects, (iii) determine a base detector non-maxima suppression (NMS) threshold to produce the highest mean mean average precision (mAP), and (iv) determine SFOD fusion pipeline’s re-scorer tuning hyperparamter to obtain the highest performance improvement
We evaluate the performance of SFOD fusion pipeline based on the mean average precision mAP according to PASCAL VOC07 challenge guidelines, using VOC07 test dataset never presented to base detectors during the training phase
Summary
Object detection popularity reflects its numerous and impactful real-world application domains, including pedestrian detection [2]–[5], manufacturing [6], [7], agriculture [8], medical [9], surveillance [10], robotics [11], security [12], object tracking and counting [13], crowd monitoring and control [14], management of emergency situations [15] and autonomous driving [16]. Regardless of the general improvements obtained from hard and soft combiners, class-conscious trainable methods reveal the highest recall performance improvement of 9.5% over the best base classifier for Android applications These ensemble methods [31], [33], [34], [36], [37] rely on specific training processes to fuse predictions of base models, leading to multiple training and tuning iterations and, eventually, to an extended deployment time. Design and implement an architecture of a non-trainable Simple Fusion of Object Detectors (SFOD) pipeline which combines predictions from base object detectors and uses a set of post-processing stages to improve the overall detection performance without requiring additional time-consuming training processes;. The combination of both similar and substantially different models offers a comprehensive analysis of the performance gains obtained from various combinations of detectors with different divergence levels
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
