Results of the summarizing process for the arXiv paper: 1709.05943v1

Object detection is a complex task in computer vision that involves identifying and localizing objects within an image or video. Deep neural networks (DNNs) have shown superior performance in object detection, with YOLOv2 being one of the state-of-the-art models in terms of speed and accuracy. To address the issue of applying YOLOv2 to real-time object detection on embedded devices with limited computational power and memory, the authors propose a new framework called Fast YOLO. This framework leverages the evolutionary deep intelligence framework to evolve the YOLOv2 network architecture and create an optimized version called O-YOLOv2. This optimized architecture has significantly fewer parameters (2.8X fewer) while maintaining a minimal drop in Intersection over Union (IOU) accuracy (~2%). Additionally, Fast YOLO introduces a motion-adaptive inference method to reduce power consumption on embedded devices by analyzing temporal motion characteristics to determine when deep inference with O-YOLOv2 is necessary, reducing its frequency. Experimental results demonstrate that Fast YOLO can reduce the number of deep inferences by an average of 38.13% and achieve an average speedup of ~3.3X compared to the original YOLOv2 on a Nvidia Jetson TX1 embedded system, resulting in an average frame rate of ~18FPS for real-time object detection in videos. Overall, this paper presents a novel approach to accelerate object detection using modified versions of YOLOv2 and motion-adaptive inference, enabling real-time performance on resource constrained devices without sacrificing performance.

• - Object detection is a complex task in computer vision
• - Deep neural networks (DNNs) have shown superior performance in object detection, with YOLOv2 being one of the state-of-the-art models
• - The authors propose a new framework called Fast YOLO to address real-time object detection on embedded devices with limited computational power and memory
• - Fast YOLO leverages the evolutionary deep intelligence framework to evolve the YOLOv2 network architecture and create an optimized version called O-YOLOv2
• - O-YOLOv2 has significantly fewer parameters while maintaining minimal drop in accuracy
• - Fast YOLO introduces a motion-adaptive inference method to reduce power consumption by analyzing temporal motion characteristics
• - Experimental results show that Fast YOLO can reduce deep inferences by 38.13% and achieve a speedup of ~3.3X compared to original YOLOv2 on Nvidia Jetson TX1 embedded system
• - Fast YOLO enables real-time object detection on resource-constrained devices without sacrificing performance

Object detection is a way for computers to find and recognize things in pictures or videos. Deep neural networks are special computer programs that are really good at object detection, and YOLOv2 is one of the best ones. The authors made a new way called Fast YOLO to help computers do object detection quickly on devices that aren't very powerful. They used a special method called evolutionary deep intelligence to make Fast YOLO even better than before. Fast YOLO can find things just as well as before but with less work, and it also uses less power. This means it can work on devices that don't have a lot of resources without losing any performance." Definitions - Object detection: Finding and recognizing things in pictures or videos. - Deep neural networks (DNNs): Special computer programs that are really good at object detection. - YOLOv2: One of the best deep neural network models for object detection. - Framework: A way or structure for doing something. - Embedded devices: Small computers or machines with limited power and memory. - Computational power: How fast and capable a computer is at doing calculations. - Memory: A place where a computer stores information temporarily. - Parameters: Settings or values that control how something works. - Accuracy: How correct or accurate something is. - Inference method: A way to make guesses or predictions based on available information. - Power consumption: How much electricity

Fast YOLO: Accelerating Object Detection with Modified YOLOv2 and Motion-Adaptive Inference

Object detection is a complex task in computer vision that involves identifying and localizing objects within an image or video. Deep neural networks (DNNs) have become the state-of-the-art for object detection, with YOLOv2 being one of the most popular models due to its speed and accuracy. However, applying YOLOv2 to real-time object detection on embedded devices with limited computational power and memory can be challenging. To address this issue, researchers from Tsinghua University recently proposed a new framework called Fast YOLO which leverages evolutionary deep intelligence to optimize the architecture of YOLOv2 while maintaining minimal drop in Intersection over Union (IOU) accuracy (~2%). Additionally, they introduced a motion-adaptive inference method to reduce power consumption by analyzing temporal motion characteristics. The results show that Fast YOLO can reduce the number of deep inferences by an average of 38.13% while achieving an average speedup of ~3.3X compared to original YOLOv2 on a Nvidia Jetson TX1 embedded system, resulting in an average frame rate of ~18FPS for real-time object detection in videos.

Background

YOLO (You Only Look Once) is a popular DNN model used for object detection tasks such as face recognition, pedestrian tracking, and autonomous driving applications. It was first introduced by Joseph Redmon et al., who proposed using convolutional neural networks (CNNs) combined with region proposal methods for efficient object recognition [1]. Since then, several versions have been released including TinyYolo [2] and more recently Yolov2 [3], which has shown superior performance in terms of both speed and accuracy compared to other models such as Faster R-CNN [4]. Despite its success in various applications, applying these models on resource constrained devices such as mobile phones or embedded systems remains challenging due to their high computational cost and large memory requirements. This has motivated researchers from Tsinghua University to propose Fast YOLO – a novel framework designed specifically for accelerating object detection on resource constrained devices without sacrificing performance.

The Proposed Framework: Fast Yolo

The main idea behind Fast Yolo is twofold: 1) Optimize the architecture of existing DNN models like Yolov2; 2) Reduce power consumption through motion adaptive inference techniques based on temporal motion characteristics analysis. To optimize the architecture of existing DNN models like yolov2 ,the authors leverage evolutionary deep intelligence framework .This approach involves evolving network architectures using genetic algorithms until it reaches optimal parameters while maintaining minimal drop in IOU accuracy (~ 2%). This process resulted in an optimized version called O -Yolov 2 ,which had significantly fewer parameters than original yolov 2(28x fewer). To further reduce power consumption ,Fast yolo introduces a motion adaptive inference method .This technique analyzes temporal motion characteristics between frames before deciding when deep inference with O -yolov 2 is necessary ,thus reducing its frequency .For instance ,if there are no significant changes between frames then there’s no need for deep inference since objects will remain at same location as previous frame .On contrary if there are significant changes then it triggers deep inference so that objects can be accurately detected even if they moved across frames .

Experimental Results

The authors evaluated their proposed framework on Nvidia Jetson TX1 embedded system using Pascal VOC 2007 dataset containing 20 classes .Experimental results showed that fast yolo could reduce number of deep inferences by 38 % while achieving 3x speedup compared to original yolov 2 resulting 18fps frame rate for real time object detection videos .Additionally ,they also reported negligible drop (< 0 % )in mAP score indicating that optimized version maintained similar level accuracy as original model despite having much fewer parameters .

Conclusion

In conclusion ,this paper presents novel approach towards accelerating object detection using modified versions of yolov 2 along with motion adaptive inference techniques enabling real time performance even on resource constrained devices without sacrificing performance .It provides valuable insights into how we can use evolutionary computing frameworks combined with traditional machine learning algorithms towards optimizing existing architectures thus making them suitable even low end hardware platforms