neural-art-mini
Lightweight version of mxnet neural art implementation using ~4.8M SqueezeNet model. Compressed model is less than 500KB
Category: Python / Deep Learning |
Watchers: 8 |
Star: 264 |
Fork: 35 |
Last update: Mar 15, 2023 |
Lightweight version of mxnet neural art implementation using ~4.8M SqueezeNet model. Compressed model is less than 500KB
hi!when i running ,there is an error: gram_loss.append(mx.sym.sum(mx.sym.square(gvar - gram[i]))) AttributeError: 'module' object has no attribute 'sum' my mxnet tag is 20160321 and neural-style is ok.
Can you tell us more information about training time and the result. I think you can update readme for showing more information
Hello. For my project I need implementation of neural style with small-size models, like this. Unfortunately, in the project we use other framework.
And now my question is : is there any mxnet-only specific staff in this implementation? Can I rewrite it with other framework (keras\tf\theano) rapidly?
P.s. also can this implementation mix to or more styles?
Thanks and best regards
I think you forget to add a parameter:init_imgae. As I know,the init image is very important for the result; It's useful for choose content,style,or random init_image for init_image So I hope oneday you can update this parameter
Would using the new SqueezeNet be faster without sacrificing quality?