YOLOv4-tiny is released

[AlexeyAB]

@hhk7734 Hi, Nice work with pypi: https://pypi.org/project/yolov4/

Did you manage to realize CIoU-loss and Mosaic?

Also YOLOv4-tiny is released:

• 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: https://github.com/AlexeyAB/darknet/issues/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.

[hhk7734]

@AlexeyAB Hey~! Hi :sunglasses:

hunglc007/tensorflow-yolov4-tflite has some problems with training. After solving this problem, I will manage BoF functions.

Then, I try to add YOLOv4-tiny. Thanks for letting me know.

[AlexeyAB]

@hhk7734 Hi,

hunglc007/tensorflow-yolov4-tflite has some problems with training.

• Did you solve this issue?

• Also how many FPS did you get for YOLO4-tiny on Coral board(TPU)? https://github.com/hhk7734/tensorflow-yolov4#objective

[hhk7734]

1. hunglc007 solved the problem, so I applied the solution.
2. Failed to compile the whole. Some sequence or operations are not compiled. The part that failed compilation was replaced by numpy. So, TPU consumes 55 ~ 60ms, the others are 35 ~ 40ms. FPS is 10~11.

[hhk7734]

CIoU

• utility: train: refactor bbox_ciou() 3a3d6522b5d59
• utility: train: move to tf.train 1bcce65e28156301
• tf: train: remove tf.math.divide_no_nan db12202497a091

Tiny

• model: yolov4: implement YOLOv4Tiny 65dcfdb33206c8710

Mosaic

• tf: dataset: implement mosaic 5c8cd15d1ccf66c0d

Requests were reflected. :)

[AlexeyAB]

@hhk7734 Great!

Do you know the reason? https://github.com/hunglc007/tensorflow-yolov4-tflite/issues/165

[hhk7734]

I studied artificial intelligence since May. So there are many things that I do not know.

[ankandrew]

@hhk7734 Very interesting to see v4 tiny on Edge TPU. I have two questions

1. What ops where not mapped to the TPU?
2. Did you quantize (post-training/training-aware) to INT8?

Thanks