Year: 2024
Type: article
Abstract: Object detection, pivotal in computer vision, spans diverse applications like autonomous driving, medical imaging, etc. Deep learning, notably, enhances detection by hierarchically representing data. Two prevalent approaches are region proposal-based (e.g., R-CNN, Fast R-CNN) and unified pipeline-based (e.g., YOLOv2). The latter, exemplified by YOLOv2, emphasizes speed and simplicity. Innovations like batch normalization and anchor boxes refine accuracy. Variants like real-time YOLO adapt for specific platforms (e.g., Non-GPU computers), while methods like SSD and DSSD optimize speed and accuracy trade-offs. Recent advancements include YOLOv3's binary cross-entropy loss for improved small object detection
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7218229269
