System information (version)
- OpenCV => 4.3.0-dev
- Operating System / Platform => Windows7 32 Bit
- Compiler => mingw64(8.1)
Detailed description
while checking out a ultra small(500k !) yolov3 model from here i got a :
cv2.error: OpenCV(4.3.0-dev) C:\p\opencv\modules\dnn\src\darknet\darknet_io.cpp:831: error: (-212:Parsing error) Unsupported activation: relu in function 'ReadDarknetFromCfgStream'
it has a couple of activation=relu lines, e.g. here but in darknet_io.cpp there is only a leaky activation:
|
if (activation == "leaky") |
|
{ |
|
setParams.setActivation("relu"); |
|
} |
calling this:
|
if (type == "relu") |
|
{ |
|
activation_param.set<float>("negative_slope", 0.1f); |
|
activation_param.type = "ReLU"; |
|
} |
so, this is indeed the "leaky" variant (used in all other yolo models), and real "relu" would have a slope of 0.0, right ?
Steps to reproduce
net = cv.dnn.readNet('yoloface-500k.cfg', 'yoloface-500k.weights')
Issue submission checklist
System information (version)
Detailed description
while checking out a ultra small(500k !) yolov3 model from here i got a :
it has a couple of
activation=relulines, e.g. here but in darknet_io.cpp there is only aleakyactivation:opencv/modules/dnn/src/darknet/darknet_io.cpp
Lines 865 to 868 in 524a2ff
calling this:
opencv/modules/dnn/src/darknet/darknet_io.cpp
Lines 223 to 227 in 524a2ff
so, this is indeed the "leaky" variant (used in all other yolo models), and real "relu" would have a slope of 0.0, right ?
Steps to reproduce
Issue submission checklist
answers.opencv.org, Stack Overflow, etc and have not found solution