Joshua Z. Zhang

Mobilenet was originally very slow on GPU, last year the depthwise convolution op has been optimized. But I am not entirely sure if CUDNN is involved in this operation.

Isn't mobilenet included in symbol_factory? https://github.com/zhreshold/mxnet-ssd/blob/master/symbol/symbol_factory.py#L132

Still working on another similar project, basically gluon detection module. I would probably come back later.

@jinfagang If you don't like iterator or you don't know how to handle that, you can refer to this tutorial, basically you are creating a module, bind it, and run...

the converter is too old and I do not have plan to fix it.

Don't measure with single image. The time spend in loading the library is different.

Are you using detector.detect to process images one by one? I admit there's more we could optimize, such as the way we load and resize the inputs.

im2rec.py was updated in mxnet. `--shuffle 1` is not supported, use `--shuffle` instead in prepare_pascal.sh

@WijdanMTak Fix the corrupted files and re-generate the record file should be fine

@WijdanMTak You can google the problem, it's due to the cuda runtime library not found in PATH.