Darknet: Do you live and breathe Yolov3?

[pfeatherstone]

@AlexeyAB Do you live and breathe Yolov3?

[AlexeyAB]

What do you mean?

[pfeatherstone]

It's just a bit of flattery. I notice that you commit regularly (almost every day), keeping up to date with the latest and greatest when it comes to object detection, inference engines and hardware platforms. Your work is valuable to a lot of people and we are all grateful. You must be very passionate about this repository.

This is of course not an issue, just a tribute.

[AlexeyAB]

Yes, it is interesting, promising, and related to my work.

The current accuracy of object detection is sufficient only for creating recommendation systems that give warnings, recommendations and tips, but accuracy is not enough for fully autonomous process control. Robots can work instead of people, and people can do intellectual work. This is potentially the largest market that has ever existed.

High accuracy of object detection required by all enterprises:

• for full automation of robots in production
• for self-driving cars
• to determine the congestion of roads, parking lots, shopping centers ...
• to detect dangerous actions

I don’t have time to do much more:

• improve object detection on video and action recognition (conv-lstm)
• object detection with ID-reidentification from differenct cameras (cosine distance between hashes)
• instant training - show the object once from different sides for training in 10 seconds
• object detection-and-tracking
• training XNOR-bit1 networks (svr)
• instance segmentation (yolact++)
• detection rotated 3d-bboxes