Related Concept Videos
Force Classification
Contact and non-contact forces are two of the most widely used categories of forces. As the name suggests, contact forces require physical contact between two objects to act upon each other. Examples of contact forces include frictional,...
Observational Learning
Difference from Background: Limit of Detection
The LOD indicates the presence or absence...
Masking and Demasking Agents
There are many masking agents, such as cyanide, fluoride, triethanolamine, thiourea, and 2,3-bis(sulfanyl)propan-1-ol (formerly 2,3-dimercapto-1-propanol), with the masking agent chosen based on...
Introduction to Learning
In contrast to learned behaviors, unlearned behaviors such as crying, sexual...
Light Acquisition
You might also read
Related Articles
Articles linked to this work by shared authors, journal, and citation graph.
Upper-Body Neuromechanical Coordination Strategies During Fatiguing Sustained Dead Hangs in Climbers.
Screen-Based Behaviors as Moderators of the Link Between Physical Activity and Internalizing and Externalizing Symptoms in Adolescents: A Population-Based Cohort Study.
Related Experiment Video
Updated: Jan 13, 2026

A Step-by-Step Implementation of DeepBehavior, Deep Learning Toolbox for Automated Behavior Analysis
Published on: February 6, 2020
Influences and Training Strategies for Effective Object Detection in Challenging Environments Using YOLO NAS-L.
Gerald Steindl1, Arnold Baca1, Philipp Kornfeind1
1Department of Biomechanics, Kinesiology and Computer Science in Sport, Centre for Sport Science and University Sports, 1150 Vienna, Austria.
Optimizing the YOLO NAS-L model for detecting small objects like swimmers requires careful tuning. Key factors for improved performance include batch accumulation, pre-trained weights, and training epochs, not just batch size or image resolution.
Area of Science:
- Computer Vision
- Machine Learning
- Object Detection
Background:
- YOLO (You Only Look Once) is a one-stage object detector.
- YOLO NAS-L is designed for enhanced small object detection.
- Detecting swimmers in aquatic environments presents a unique challenge.
Purpose of the Study:
- To investigate the impact of dataset characteristics, training strategies, and hyperparameters on YOLO NAS-L performance.
- To optimize YOLO NAS-L for detecting swimmers in challenging aquatic environments.
- To identify critical factors for efficient development of high-performing YOLO NAS-L models.
Main Methods:
- Systematic investigation of YOLO NAS-L performance.
- Evaluation using mean Average Precision (mAP) and F1-score.
- Analysis of hyperparameters: batch size, batch accumulation, epochs, image resolution, pre-trained weights, and data augmentation.
Main Results:
- Batch size and image resolution showed limited impact on performance.
- Batch accumulation, pre-trained weights, and training epochs were critical for optimization.
- Optimized hyperparameters and training strategies significantly improved model performance.
Conclusions:
- Effective YOLO NAS-L model development relies on a combination of optimized hyperparameters and training strategies.
- Pre-trained weights are crucial for achieving high performance in object detection tasks.
- Careful tuning of training epochs is essential for maximizing model effectiveness.
