A new version of YOWOv2 is being developed by us.

[Hope1337]

Update : YOWOv3 paper has been public here
Official code has been public

First of all, I would like to express my sincere gratitude to the author. This work is truly amazing and helpful. This repo has been tremendously helpful to me during my research in the field of spatiotemporal motion detection. In the past few months, I have frequently reviewed the code, and today I noticed that the author has updated the README.

It's unfortunate that the author has stopped providing support for the issues in this repo. However, I am also conducting my own research on a YOWO version and have achieved some promising results:

UCF-101: 84.x mAP with a 54M parameter model
AVAv2.2: 20.21 mAP with a 96M parameter model
Update : UCF-101 : 88.24 mAP with a 59.8M #param model !
Update : AVAv2.2 : 20.31 mAP with a 59.86M #param model !

I have just made my code public at : https://github.com/AakiraOtok/Project_VU

My research focuses on utilizing different techniques to improve the YOWOv2 model, along with support for various experimental configurations. The usage instructions for the repo are still quite sparse at the moment because I am still fine-tuning my model. However, they will be updated soon if people are interested. And of course, I will also provide support and answer any questions regarding my repo.

Once again, thank you to the author @yjh0410.