search

Adamdad / keras-YOLOv3-mobilenet

I transfer the backend of yolov3 into Mobilenetv1，VGG16，ResNet101 and ResNeXt101
MIT License
563 stars 166 forks source link

advantage of converting the Darknet-model to a Keras-model #13

Open phamngocthanhtrung opened 5 years ago

phamngocthanhtrung commented 5 years ago

Hi, what is the advantage of converting the Darknet-model to a Keras-model ? Best Regards

Adamdad commented 5 years ago

the structure is the same, but keras is easier to understand and wide-used

phamngocthanhtrung commented 5 years ago

the structure is the same, but keras is easier to understand and wide-used

yes thank you, and the runtime remains the same?

Adamdad commented 5 years ago

No, because darknet is based on pure c, so darknet is faster, but hard to comprehend. Darknet is faster than tensorflow and keras.

phamngocthanhtrung commented 5 years ago

it make sense! one last question please! if we have the same structure, does it mean we have the same prediction (MAP/IoU) ?

Adamdad commented 5 years ago

That‘s not always true. Different Deep learning Framework use different tricks and skills and on their implementation. For example, the parameter initialization is very much important for the precision of CNN model, with different frameworks using different strategies.