Joshua Z. Zhang

Good point! I guess I made a mistake cause I am completely disordered by in-place operation like activations. I will try to test the performance by fixing these errors. Thanks...

I've changed the conv to activation output, still got very similar results. I need more spare time to debug it.

@liumusicforever Exactly

@ZongweiZhou1 It's a warm up from very small learning rate to normal in burn_in period.

Hang tight, I will be investigate and implement this in a more convenient package. Will inform once finished.

Yolo3 is ongoing in gluon cv package: https://github.com/dmlc/gluon-cv

@prnvjb Please checkout https://gluon-cv.mxnet.io/model_zoo/index.html#object-detection, tons of best models are available now for detection!

it's difficult object if 1

Measure single image is not accurate especially on GPU, because cuda/mxnet engine need to be cold initialized. You can try batching a series of images, or even dummy duplicate images...

#13 Is a nice example