YOLOv4-tiny released

[AlexeyAB]

Feature-request: YOLOv4-tiny (detector)

• source code: https://github.com/AlexeyAB/darknet
• cfg: https://raw.githubusercontent.com/AlexeyAB/darknet/master/cfg/yolov4-tiny.cfg
• weights: https://github.com/AlexeyAB/darknet/releases/download/darknet_yolo_v4_pre/yolov4-tiny.weights
• discussion: YOLOv4-tiny released: 40.2% AP50, 371 FPS (GTX 1080 Ti), 1770 FPS tkDNN/TensorRT AlexeyAB/darknet#6067

There is required only 1 feature:

1. Add groups= and group_id= to the [route] layer.

[route]
layers=-1
groups=2
group_id=1

So if input is WxHxC, it divides input into 2 groups WxHx(C/2) (there are 2 groups: 0 and 1), and loads the 2nd group_1 WxHx(C/2).

If there are many layers specified in layers= parameter, then this will be done for each of the input layers specified in layer=, then results will be concatenated across channels.

[ersheng-ai]

Thank you for your useful information

[Algolzw]

I can't use yolov4-tiny in this PyTorch repository, is there any solution?

[ersheng-ai]

I can't use yolov4-tiny in this PyTorch repository, is there any solution?

This repository is not designed specifically for yolov4-tiny.
You can imitate implementations here to run yolov4-tiny but you have to:

1. re-write pytorch definitions of nn modules in models.py
2. Update anchor arguments or definitions for yolo layer to suite yolov4-tiny

Maybe there will be support of yolov4-tiny here in the future.