Darknet: SAM in yolo v4 use sigmoid or mish?

Created on 27 Apr 2020  路  9Comments  路  Source: AlexeyAB/darknet

According to the Fig.5 in paper, it shows that using sigmoid in SAM block, but in the config file it seems that mish is used.
https://github.com/AlexeyAB/darknet/blob/6033baabc9d646b23d8dbdb21cf84663910dca60/cfg/yolov4-custom.cfg#L122

All 9 comments

Hello,

Provided yolov4.cfg does not have SAM block.
image

an example of SAM block is as follow:

[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=512
activation=mish

[convolutional]
batch_normalize=1
filters=512
size=1
stride=1
pad=1
activation=logistic

[sam]
from=-2

thank you so is this part in cfg is WRC?

No, Mish is just an activation function.

For more information about WRC, you can take a look https://github.com/AlexeyAB/darknet/issues/4662.

I know mish is an activation.
I compared yolov4 cfg with yolov3 ,and found there are some different in backbone's structure. If those diff is not WRC, is it CSP?
Besides, modified pan seems not in this cfg as well.

thank you so much

Hi @AlexeyAB where to find this cfg file yolov4-sam-mish.cfg

Hi @WongKinYiu Thanks. I would like to ask how to implement conv_lstm in yolov4?

Was this page helpful?
0 / 5 - 0 ratings

Related issues

louisondumont picture louisondumont  路  3Comments

hemp110 picture hemp110  路  3Comments

Jacky3213 picture Jacky3213  路  3Comments

qianyunw picture qianyunw  路  3Comments

shootingliu picture shootingliu  路  3Comments