opconty / keras-shufflenetv2 Goto Github PK

Hello !

I am having a lot of errors when i try to make this compatible with the latest Keras & Tensorflow 2.0 .
I have to make the training on Kaggle and then to save the model.h5 and use it on Raspberry Pi 4.

Can you edit the code to be able to run with Keras & Tensorflow latest updates?

Here you can find the link to Kaggle: https://www.kaggle.com/dcosmin/shufflenet-with-keras.

Thank you !
I really need this ShuffleNet Model for a project !

Hi, I downloaded your pre-trained weights and tried to load it, but I got error like:

"ValueError: You are trying to load a weight file containing 31 layers into a model with 111 layers."

model = ShuffleNetV2(include_top=True）
model.load_weights('shufflenetv2_emotion_recogn.h5')

Could you please tell me how to resolve it?

Dear opconty,
Thanks for sharing your codes firstly. Of course, I've met some problems when I want to use your codes. Hope that you can help me!
Where is the main function to train this network?
I've added this shufflenetV2 network into my own objects, but there were always some errors. So, I want to know how do you add this shuffle network into your object. Would you mind sharing your train.py (or something like this) to me?

stage2, output_channels如果是116, 这个代码那两条支路输出都是116，concat后stage2就是232了吧？
(If at stage2, the output_channels is 116, this function will output 232 channels after concat.)

Hi,thanks for your work, I have one question, for norm and mean values, is same as PyTorch?

mean = [0.485, 0.456, 0.406] and std = [0.229, 0.224, 0.225]

https://pytorch.org/hub/pytorch_vision_shufflenet_v2/

… or what values I need to use?

I follow this tutorial for this

https://www.kaggle.com/dcosmin/shufflenet-with-keras/comments

thanks

x = K.reshape(x, [-1, height, width, 2, channels_per_split])
#Failed to convert object of type <class 'list'> to Tensor. Contents: [-1, None, None, 2, 16]. Consider casting #elements to a supported type.

Hello, thanks for you sharing.
I have the question, is there a model pre-trained on ImageNet Datasets provided or should I train from scratch?
Thanks & Regards!