What is the latest version of YOLO? Is V5 a scam?

[franva]

Hi guys,

I am learning the YOLO, it looks great~! I think this Github repo and this website are the official websites for YOLO?

But recently I saw an article talking about YOLO V5.

My 1st thought is : what? Since what time does YOLO get V5??

I searched from the Cornell University and could not find anything about YOLO V5 either.

1. Could someone tell me what is the latest version of YOLO?
2. What is the official website for YOLO?

Appreciated,

Winston

[AlexeyAB]

The latest version - YOLOv4 (YOLOv4 and Scaled-YOLOv4), with paper, with URLs from official repository, and with the best Accuracy/Speed among all known algorithms.

YOLOv5-Ultralytics - model is worse than Scaled-YOLOv4, without a scientific article.

• Negative opinion of Deep Learning developers about Yolov5-Ultralytics: https://www.reddit.com/r/MachineLearning/comments/h0ddia/news_yolov5_is_here_stateoftheart_object/

• Official repository - declares the latest version of YOLOv 4 / Scaled-YOLOv4: https://github.com/pjreddie/darknet#darknet

• Official YOLOv4 repository: https://github.com/AlexeyAB/darknet

• Official paper - YOLOv4: https://arxiv.org/abs/2004.10934

• Official paper - Scaled-YOLOv4: https://arxiv.org/abs/2011.08036

• Taiwanese government uses YOLOV4: https://www.taiwannews.com.tw/en/news/3957400

• Amazon implemented DistanceAssistant on ROS by using YOLOv4: https://github.com/amzn/distance-assistant

• BMW uses YOLOv4: https://github.com/BMW-InnovationLab in several projects training, inference_GPU, inference_CPU

• Official site (v3 is Joseph’s latest version of Yolo, after which Joseph stopped developing Yolo): https://pjreddie.com/darknet/yolo/

• Amazon, Microsoft, Intel and Apple wanted to buy the Xnor AI company (the company develops Yolo): https://www.geekwire.com/2020/seattle-ai-startup-drew-interest-amazon-microsoft-intel-selling-apple/

• Xnor AI (the company develops Yolo) was bought by Apple for $200 M: https://www.forbes.com/sites/janakirammsv/2020/01/19/apple-acquires-xnorai-to-bolster-ai-at-the-edge/#20a12e943975

• All four registered trademarks (YOLO9000, YOLOV2, YOLOV3, TINY YOLO) have officially been transferred to Apple: https://www.patentlyapple.com/patently-apple/2020/06/four-xnorai-trademarks-covering-yolo-tools-for-real-time-object-detection-have-officially-been-transferred-to-apple.html

---

• The official statement that the new Yolo repository is https://github.com/AlexeyAB/darknet https://twitter.com/pjreddie/status/1253891078182199296

---

• YOLOv5 is worse than YOLOv4:

---

AP50:95 - FPS (Tesla V100) Paper: https://arxiv.org/abs/2011.08036

• Comparison of YOLOv4 with other models: https://github.com/AlexeyAB/darknet

---

• YOLOv4-tiny released: 40.2% AP50, 371 FPS (GTX 1080 Ti) / 330 FPS (RTX 2070) https://github.com/AlexeyAB/darknet/issues/6067

---

• Speed (FPS) of YOLOv4 in different libraries with different batch-size - up to 400 FPS:

---

• Governments use YOLOv4 in real projects: https://www.sinica.edu.tw/en and https://www.taiwannews.com.tw/en/news/3957400 YOLOv4 has been used in the development of “Smart City Traffic Flow Solutions”, a collaborative project with ELAN Microelectronics Corporation to enhance smart city innovation in Taiwan. But also it can be used for the development of self-driving vehicles, the analysis of medical images, and the testing of faulty equipment in factories.

[franva]

Firstly, thank you for your detailed explanation about where YOLO comes from, who firstly invented YOLO and how YOLO is currently going. This really clears my mind about the history of YOLO and who is the official and who is not, so thumb up for this :)

After knowing these information, as a beginner of AI I can tell why the article I referenced in my question is so confusing.

In the video mentioned in the article, they claimed that their fake YOLOV5 aiming at different perspectives ,e.g. ease of use, exportability, memory requirement, etc...

With not much AI background knowledge, as a beginner, I don't think their fake YOLOV5 should use the name YOLO at all. If a new model architecture is created, then I would expect to see its related academic papers on some scientific websites. If it's just a tweak of an existing YOLO V3, the author of fake YOLO V5 should name it something like YOLOV3-PyTorch.

As a beginner of AI, I dislike this kind of behavior, as it is waste of my time to learn such a fake architecture and I am sure there are many beginners who will have similar experience as mine, but not all of them have luck to discovered that "YOLOV5" is a fake one. So as an owner of YOLO V4, and the owner of YOLO V3 and all other owners of YOLO should do something to make people aware of this wrongdoing.

So here is my thought, I feel that many of users(developers) will be familiar with Python and PyTorch framework based AI models versus C based YOLO models. This is where the selling point of the fake YOLO V5 is. Therefore, why not the owner(@AlexeyAB ) of this YOLO V4, create an authentic YOLO V4 for PyTorch(Python) version model? I feel this is a better way to defeat the fake one.

[AlexeyAB]

I am not opposed the Ultralytics repository. I am opposed to unfair comparisons with YOLOv4. YOLOv4 can be trained on both repos: https://github.com/ultralytics/yolov3 and https://github.com/ultralytics/yolov5 So in the future, a more accurate and faster version of YOLO on these repositories may indeed be released. And it will be very convenient for those who want to use Python and Pytorch.

[AlexeyAB]

YOLOv4 training and inference on different frameworks / libraries:

Pytorch-implementations:

• https://github.com/WongKinYiu/PyTorch_YOLOv4
• https://github.com/Tianxiaomo/pytorch-YOLOv4
• 3D-rotated-bboxes: https://github.com/maudzung/Complex-YOLOv4-Pytorch

TensorFlow: https://github.com/hunglc007/tensorflow-yolov4-tflite

OpenCV (YOLOv4 built-in OpenCV): https://github.com/opencv/opencv

TensorRT: https://github.com/ceccocats/tkDNN

Tencent/NCNN: https://github.com/Tencent/ncnn

TVM https://tvm.ai/about

OpenDataCam: https://github.com/opendatacam/opendatacam#-hardware-pre-requisite

BMW-InnovationLab - Training with YOLOv4 has never been so easy (monitor it in many different ways like TensorBoard or a custom REST API and GUI):

• Training: https://github.com/BMW-InnovationLab/BMW-YOLOv4-Training-Automation
• Inference on GPU: https://github.com/BMW-InnovationLab/BMW-YOLOv4-Inference-API-GPU
• Inference on CPU: https://github.com/BMW-InnovationLab/BMW-YOLOv4-Inference-API-CPU