COVID-19 Cases
Newserector
2020-06-01

YOLO High Speed Object Detection In Real Time

Image:-Video

Just once (YOLO) you look, it's a state of the art object detection system in real time. On the Pascal Titan X, pictures at 30 FPS are processed and COCO test-dev has a mAP of 57.9 percent.

View this post on Instagram

A new version of the popular YOLO (You Only Look Once) has just been released less than a month ago! It more accurate and way faster than all the current object detection algorithms and you can train it and use it in your home with a single average GPU! I introduced it in a video that you can see here if you want to learn more about it: https://youtu.be/CtjZFkO5RPw The link is in my bio as well. . . . . . #ai #OnlineLearning #machinelearning #computerscience #programmer #software #datascience #tech #code #coder #softwaredeveloper #design #technology #pythonprogramming #business #startup #data #neuralnetworks #artificialintelligence #deeplearning #bigdata #developer #coding #datascientist #python

A post shared by What's Artificial Intelligence (@whats_ai) on

In the last few years even the detection of the object begins to mature. As illustrated below, YOLOv4 claims to have the latest precision, while maintaining a high frame rate for processing. The accuracy of the MS COCO with an inferior speed of around 65 FPS at Tesla V100 is 43.5 percentAP (65.7 percentAP50). High precision is no longer the only sacred grail in object detection. We want the model in the edge devices to work smoothly. It is also important how to process video input in real time with inexpensive hardware.

The fun reading of YOLOv4 is the evaluation, modification and integration of new technologies in YOLOv4. And it changes also in order to make the detector more suitable for a single GPU training.

YOLOV4 is optimal for detecting objects in real time because the network is optimal in Pareto

YOLOv3 is very quick and precise. In mAP with .5 IOU YOLOv3, the focus loss is approximately equivalent but 4 times faster. Moreover, by simply changing the model size, you can easily compromise between speed and accuracy without any reworkout!

Check out the github code

Article Edited by | Jhon H |

we hope you're doing well if you see any inappropriate phrases please let us know on our contact page at the bottom. thank you!    .

320

Posts

3.8

Rating

3.2

Points