VGG19 Implementation with example

[OwenLiuzZ]

Checklist

• I've tested that my changes are compatible with the latest version of Tensorflow.
• I've read the Contribution Guidelines
• I've updated the documentation if necessary.

Motivation and Context

Description

[DEKHTIARJonathan]

@OwenLiuzZ Thank you so much for your PR.

Please update the changelog with your contribution and add your name to the contributor list.
Please follow the formating

I let @zsdonghao reviewing your work ;)

[zsdonghao]

Hi, please also update the docs https://github.com/tensorlayer/tensorlayer/blob/master/docs/modules/models.rst

and the changelog https://github.com/tensorlayer/tensorlayer/blob/master/CHANGELOG.md

[zsdonghao]

A better coding style can be found here https://github.com/tensorlayer/tensorlayer/blob/master/tensorlayer/models/squeezenetv1.py, instead of using self.all_layers = net.xxxx, we set the model as Layer class.
Therefore, for other models, you can connect tfslim into the layer class.

[zsdonghao]

we should make it download automatically. @lgarithm @luomai any idea?

[OwenLiuzZ]

One simple solution is that we train all the models and store theses models in the libraryof the TL

[luomai]

I got a "This site can’t be reached" error when I tried to download the npy of VGG'19 from the website given.

How large is the npy file? Can we host it in our pre-trained model repo?

[OwenLiuzZ]

Yes, it is about 570MB for this npy file. The public repo maximum limit for the file is 100 MB. However, I do not know what the limit is in the private repo.

[lgarithm]

The original download link is for browser download only.
I've uploaded it to tensorlayer/pretrained-models, you can now use this download link in the program:

https://media.githubusercontent.com/media/tensorlayer/pretrained-models/master/models/vgg19.npy

[lgarithm]

You can use ‘git lfs’ for uploading large file.

________________________________ From: Sichao Liu <notifications@github.com> Sent: Thursday, June 14, 2018 5:10:38 PM To: tensorlayer/tensorlayer Cc: LG; Mention Subject: Re: [tensorlayer/tensorlayer] [WIP] VGG19 Implementation with example (#698) @OwenLiuzZ commented on this pull request.

________________________________ In tensorlayer/models/vgg19.py<#698 (comment)>:

+ lambda net: MaxPool2d(net, filter_size=(2, 2), strides=(2, 2), padding='SAME', name='pool5'),

+ lambda net: FlattenLayer(net, name='flatten'), + lambda net: DenseLayer(net, n_units=4096, act=tf.nn.relu, name='fc1_relu'), + lambda net: DenseLayer(net, n_units=4096, act=tf.nn.relu, name='fc2_relu'), + lambda net: DenseLayer(net, n_units=1000, act=tf.identity, name='fc3_relu'), + ] + net = net_in + for l in layers: + net = l(net) + # if end_with in net.name: + if net.name.endswith(end_with): + return net + + raise Exception("unknown layer name (end_with): {}".format(end_with)) + + def restore_params(self, sess): Yes, it is about 570MB for this npy file. The public repo maximum limit for the file is 100 MB. However, I do not know what the limit is in the private repo. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<#698 (comment)>, or mute the thread<https://github.com/notifications/unsubscribe-auth/AB8ydOuPFO-sYon3M0JU9R09BJHzIG9yks5t8iiOgaJpZM4Um-Q1>.

[lgarithm]

@zsdonghao for vgg16 we used the pre-trained weights provided by http://www.cs.toronto.edu/~frossard/post/vgg16/.
But there is no corresponding weights for vgg19.
Where can we find a publicly available pre-trained vgg19 model for TF/TL?

[DEKHTIARJonathan]

@lgarithm https://github.com/tensorflow/models/tree/master/research/slim

[OwenLiuzZ]

@lgarithm I use the pre-trained model from this repo

[lgarithm]

Hi, @OwenLiuzZ I've tested this PR on a clean VM, it works!
Please resolve conflicts and then it will be ready to merge.

[OwenLiuzZ]

Hi, @lgarithm I have resolve the conflicts. Thanks for reminding.