Nguyễn Duy Cương

[Update] The caffe converter tool has issue with depthwise convolution layer in mobilenet. I fixed it manually and i could run mobilenet_ssd_300 model in Mxnet now with mAP=70.9% on VOC...

@madhavajay you can use mobilenet_ssd_300 models in [my repo](https://github.com/titikid/DeepLearning)

Thanks @zhreshold , it worked!

@zhreshold is this also fix 'gamma' term? Should i remove 'fix_gamma=True' in bn layer?

I already trained 2 models from scratch, all parameters is set as default (lr=0.004, batch=48, single gpu) - model with fixed beta (only in base mobilenet network) and gamma: ~41.5%...

Hi @zhreshold I found that if i remove "beta" term only, the convolution still has a small shift factor because the impact of "running_mean" term. i set "lr_mult" of "running_mean"...

@zhreshold can you give me some suggestion?

@zhreshold i'm not really clear what do you mean for now, but i will investigate it. Thanks!

just rename extract_features.bin --> extract_features in your script!

i wonder how this loss calculated? for all training samples or only small part (mini-batch) of them?