Improved Hard Example Mining Approach for Single Shot Object Detectors

Köksal, Aybora
Tuzcuoğlu, Önder
Ince, Kutalmiş Gökalp
Ataseven, Yoldaş
Alatan, Abdullah Aydın
Hard example mining methods generally improve the performance of the object detectors, which suffer from imbalanced training sets. In this work, two existing hard example mining approaches (LRM and focal loss, FL) are adapted and combined in a state-of-the-art real-time object detector, YOLOv5. The effectiveness of the proposed approach for improving the performance on hard examples is extensively evaluated. The proposed method increases mAP by 3% compared to using the original loss function and around 1-2% compared to using the hard-mining methods (LRM or FL) individually on 2021 Anti-UAV Challenge Dataset.
2022 IEEE International Conference on Image Processing (ICIP)


Comparison of Equivalent Stress Methods with Critical Plane Approaches for Multiaxial High Cycle Fatigue Assessment
Engin, Zafer; Çöker, Demirkan (Elsevier BV; 2017-09-07)
Several equivalent stress methods and more advanced critical plane criteria are compared in terms of their performance in fatigue life estimations under uniaxial and biaxial loadings in which the effect of phase is investigated. For this purpose a MATLAB code is written which transforms the multiaxial cyclic stress state into a uniaxial cyclic stress to use with the equivalent stress based methods. For critical plane approaches the program searches a damage parameter on all material planes. The maximum dama...
Investigating the correlation between magnetic Barkhausen noise emission and the fatigue life of shot-peened AISI 4140 steel
Çalışkan, S.; Gür, Cemil Hakan (British Institute of Non-Destructive Testing (BINDT), 2019-01-01)
The performance of engineering components under conditions of fatigue can be improved by shot peening. The location and magnitude of the highest residual compressive stress, the depth of the compression zone and the stability of the residual stress state during cyclic loading are critical parameters for maximum utilisation of the advantageous influence of shot peening on fatigue performance. For accurate assessment of fatigue lifetime, residual stresses should be considered because they are algebraically su...
A stochastic programming approach for the disassembly line balancing with hazardous task failures
Kalaycilar, Eda Goksoy; Batun, Sakine; Azizoğlu, Meral (2021-04-01)
We consider a partial disassembly line balancing problem with hazardous tasks whose successful completions are uncertain. When any hazardous task fails, it causes damages of the tasks on the workstation that it is performed on and all remaining tasks to be performed in the succeeding workstations. We attribute probabilities for the successful completion and failure of the hazardous tasks and aim to maximise the total expected net revenue. We formulate the problem as a two-stage stochastic mixed-integer prog...
Basic requirement for modelling fluidized beds: Fast computation of particle size distributions (PSDs)
Selçuk, Nevin; Degirmenci, E (1996-06-01)
A fast computation of PSDs in fluidized beds is a basic requirement for the mathematical modelling of the behaviour of fluidized beds as it is the most time consuming step in the complete prediction procedures. In an attempt to remove this bottleneck, a computationally efficient approach for the solution of population balance of Kunii and Levenspiel is developed. This new approach replaces the conventional numerical integration of the analytical solution of the population balance with direct integration in ...
Evaluation of Ground motion Record Scaling and Selection Methods for Seismic Design and Assessment of Structures
Ay, Bekir Özer; Crowley, Helen (null; 2015-10-16)
Amplitude scaling and selection of ground-motion records are frequently used tools to obtain a group of accelerograms suitable for structural performance assessment for a given hazard level. To provide a suitable set of records to be used in nonlinear dynamic analysis either to estimate median structural response or to predict its full probability distribution, “n” scaled ground motions that best match with target spectrum are selected. Common concerns related with this approach are the use of appropriate d...
Citation Formats
A. Köksal, Ö. Tuzcuoğlu, K. G. Ince, Y. Ataseven, and A. A. Alatan, “Improved Hard Example Mining Approach for Single Shot Object Detectors,” presented at the 2022 IEEE International Conference on Image Processing (ICIP), Bordeaux, Fransa, 2022, Accessed: 00, 2022. [Online]. Available: