YOLOv4

[noobgrow]

Dear sir,
are you working on a new version like YOLOv4? Some new methods such as CornetNet-Lite are reported faster and have highest mAP. Will you beat them back ?

[AlexeyAB]

@noobgrow Hi,

I will think about implementing some ideas from CornerNet/Lite to the new Yolo models:

• CornerNet: Detecting Objects as Paired Keypoints: https://arxiv.org/abs/1808.01244v2

• CornerNet-Lite: Efficient Keypoint Based Object Detection: https://arxiv.org/abs/1904.08900v1

---

I am currently developing: https://github.com/AlexeyAB/darknet/projects/1

something like:

• conv-LSTM network https://arxiv.org/abs/1903.10172 the best possible mAP/speed even on SmartPhone

• and PANnet ( the 1st place in the COCO 2017 Challenge Instance Segmentation task and the 2nd place in Object Detection task without large-batch training.) https://arxiv.org/abs/1803.01534v4

[noobgrow]

OK，thanks for your sharing，So you are trying to add something like mask branch and fast-slow structrue for video? And have u read this, https://arxiv.org/abs/1904.02689, one stage segmentation method
I think what you are doing is really meaningful. In fact, that's what I want to improve. but I'm not so good at coding, so I really looking forward to your work, where can I get the latest news about your work, or is there a deadline for you?

[AlexeyAB]

@noobgrow

OK，thanks for your sharing，So you are trying to add something like mask branch and fast-slow structrue for video?

Not, yet: #3114 (comment)

---

And have u read this, https://arxiv.org/abs/1904.02689, one stage segmentation method
I think what you are doing is really meaningful.

Yes, I read: #3048

---

In fact, that's what I want to improve. but I'm not so good at coding, so I really looking forward to your work, where can I get the latest news about your work, or is there a deadline for you?

Some plans: https://github.com/AlexeyAB/darknet/projects