shicai comments

Results 72 comments of


                                            shicai

Could you share your detailed training settings ?

please refer to: https://github.com/shicai/MobileNet-Caffe/issues/1

can not load pretrained model with the deploy prototxt

are you using NVIDIA caffe?

can not load pretrained model with the deploy prototxt

there are only 3 blobs for batch norm layers. please see: https://github.com/BVLC/caffe/blob/master/src/caffe/layers/batch_norm_layer.cpp

can not load pretrained model with the deploy prototxt

please check the path of your caffe, and the codes of BN layers.

Not getting a good acc

what is your result? solver prototxt file has no effect on the val accuracy of the pretrained model.

Not getting a good acc

training models on ImageNet is somehow very tricky. you'd better train some simple models first.

Not getting a good acc

please see: https://github.com/shicai/MobileNet-Caffe/issues/1

keras model to caffe

you need to write your own api to convert keras models into caffe format.

something went wrong with image preprocessing and post-processing. - add `transformer.set_input_scale('data', 0.017)` - add these codes to the end of your prototxt: `layer { name: "prob" type: "Softmax" bottom: "fc7"...

Fail to run on GPU

I am not sure what the problem is. have you changed the kernel numbers for some layers?