VGG-16 pre-trained model for Keras

readme.md

##VGG16 model for Keras

This is the Keras model of the 16-layer network used by the VGG team in the ILSVRC-2014 competition.

It has been obtained by directly converting the Caffe model provived by the authors.

Details about the network architecture can be found in the following arXiv paper:

Very Deep Convolutional Networks for Large-Scale Image Recognition
K. Simonyan, A. Zisserman
arXiv:1409.1556

In the paper, the VGG-16 model is denoted as configuration D. It achieves 7.5% top-5 error on ILSVRC-2012-val, 7.4% top-5 error on ILSVRC-2012-test.

Please cite the paper if you use the models.

###Contents:

model and usage demo: see vgg-16_keras.py

weights: vgg16_weights.h5

vgg-16_keras.py

from keras.models import Sequential
from keras.layers.core import Flatten, Dense, Dropout
from keras.layers.convolutional import Convolution2D, MaxPooling2D, ZeroPadding2D
from keras.optimizers import SGD
import cv2, numpy as np

def VGG_16(weights_path=None):
    model = Sequential()
    model.add(ZeroPadding2D((1,1),input_shape=(3,224,224)))
    model.add(Convolution2D(64, 3, 3, activation='relu'))
    model.add(ZeroPadding2D((1,1)))
    model.add(Convolution2D(64, 3, 3, activation='relu'))
    model.add(MaxPooling2D((2,2), strides=(2,2)))

    model.add(ZeroPadding2D((1,1)))
    model.add(Convolution2D(128, 3, 3, activation='relu'))
    model.add(ZeroPadding2D((1,1)))
    model.add(Convolution2D(128, 3, 3, activation='relu'))
    model.add(MaxPooling2D((2,2), strides=(2,2)))

    model.add(ZeroPadding2D((1,1)))
    model.add(Convolution2D(256, 3, 3, activation='relu'))
    model.add(ZeroPadding2D((1,1)))
    model.add(Convolution2D(256, 3, 3, activation='relu'))
    model.add(ZeroPadding2D((1,1)))
    model.add(Convolution2D(256, 3, 3, activation='relu'))
    model.add(MaxPooling2D((2,2), strides=(2,2)))

    model.add(ZeroPadding2D((1,1)))
    model.add(Convolution2D(512, 3, 3, activation='relu'))
    model.add(ZeroPadding2D((1,1)))
    model.add(Convolution2D(512, 3, 3, activation='relu'))
    model.add(ZeroPadding2D((1,1)))
    model.add(Convolution2D(512, 3, 3, activation='relu'))
    model.add(MaxPooling2D((2,2), strides=(2,2)))

    model.add(ZeroPadding2D((1,1)))
    model.add(Convolution2D(512, 3, 3, activation='relu'))
    model.add(ZeroPadding2D((1,1)))
    model.add(Convolution2D(512, 3, 3, activation='relu'))
    model.add(ZeroPadding2D((1,1)))
    model.add(Convolution2D(512, 3, 3, activation='relu'))
    model.add(MaxPooling2D((2,2), strides=(2,2)))

    model.add(Flatten())
    model.add(Dense(4096, activation='relu'))
    model.add(Dropout(0.5))
    model.add(Dense(4096, activation='relu'))
    model.add(Dropout(0.5))
    model.add(Dense(1000, activation='softmax'))

    if weights_path:
        model.load_weights(weights_path)

    return model

if __name__ == "__main__":
    im = cv2.resize(cv2.imread('cat.jpg'), (224, 224)).astype(np.float32)
    im[:,:,0] -= 103.939
    im[:,:,1] -= 116.779
    im[:,:,2] -= 123.68
    im = im.transpose((2,0,1))
    im = np.expand_dims(im, axis=0)

    # Test pretrained model
    model = VGG_16('vgg16_weights.h5')
    sgd = SGD(lr=0.1, decay=1e-6, momentum=0.9, nesterov=True)
    model.compile(optimizer=sgd, loss='categorical_crossentropy')
    out = model.predict(im)
    print np.argmax(out)

@zero-craft by_name=True simply looks for layers with names of layers in the model, and tries to cast those same layers from the weights file. This circumvents the error but essentially initializes the model with random weights.

Has anyone been able to get a solution to :
You are trying to load a weight file containing 0 layers into a model with 16 layers.

Loading the model from github is extremely slow (in my case). Kaggle also has a mirror: https://www.kaggle.com/keras/vgg16#vgg16_weights_tf_dim_ordering_tf_kernels.h5

Hi
My database is Fashion-mnist, and the size of my photos is 28 * 28. How can I use vgg architecture? Given that the default input size for this model is 224x224

@mahboubeh86

So if I remember correctly the VGG model can take at minimum 32x32 input. So what I would do is pad your images with 2 blank pixels all around. Then when you load the vgg model you can simply pass in the parameter input_size.

Check out the code here: https://github.com/keras-team/keras-applications/blob/master/keras_applications/vgg19.py

hi Thanks for your help. I read the link to the code you sent, and if I want to use it my whole program will change. If you can send me a resize code sample. Because my changes are associated with a VGG16 architecture error. thanks

…

On Mon, Dec 16, 2019 at 11:37 PM Njerschow ***@***.***> wrote: @mahboubeh86 <https://github.com/mahboubeh86> So if I remember correctly the VGG model can take at minimum 32x32 input. So what I would do is pad your images with 2 blank pixels all around. Then when you load the vgg model you can simply pass in the parameter input_size. Check out the code here: https://github.com/keras-team/keras-applications/blob/master/keras_applications/vgg19.py — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <https://gist.github.com/07d7802847aaad0a35d3?email_source=notifications&email_token=AOBV3GEG3Z3PJUCJQAHVZ4TQY7NW5A5CNFSM4HHECR5KYY3PNVWWK3TUL52HS4DFVNDWS43UINXW23LFNZ2KUY3PNVWWK3TUL5UWJTQAF6AQA#gistcomment-3113216>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AOBV3GEST3NWJRE3XHO5JADQY7NW5ANCNFSM4HHECR5A> .

can you show me how you are using the code in your program? Are you getting an error when you run it?

Hi yes, i'll send you the code, when i want to change dimensions, error code, and when i change the dimensions of vgg architecture, error architecture. Thanks

…

On Thu, 19 Dec 2019, 22:08 Njerschow, ***@***.***> wrote: can you show me how you are using the code in your program? Are you getting an error when you run it? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <https://gist.github.com/07d7802847aaad0a35d3?email_source=notifications&email_token=AOBV3GCYWT6AJAAWEHACS2LQZO5SZA5CNFSM4HHECR5KYY3PNVWWK3TUL52HS4DFVNDWS43UINXW23LFNZ2KUY3PNVWWK3TUL5UWJTQAF6G2E#gistcomment-3116450>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AOBV3GGI36IIEYBK2G6VU33QZO5SZANCNFSM4HHECR5A> .

hi Can you please take the time to let me know if you see the code? And help me write code for resizing and identifying it with the vgg16 architecture. Thankful

…

On Thu, Dec 19, 2019 at 10:08 PM Njerschow ***@***.***> wrote: can you show me how you are using the code in your program? Are you getting an error when you run it? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <https://gist.github.com/07d7802847aaad0a35d3?email_source=notifications&email_token=AOBV3GCYWT6AJAAWEHACS2LQZO5SZA5CNFSM4HHECR5KYY3PNVWWK3TUL52HS4DFVNDWS43UINXW23LFNZ2KUY3PNVWWK3TUL5UWJTQAF6G2E#gistcomment-3116450>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AOBV3GGI36IIEYBK2G6VU33QZO5SZANCNFSM4HHECR5A> .

I dont see your code @mahboubeh86

hi Yes, I'm going to work with the fashionmnist data set and vgg16 architecture , and I'm trying to convert the fashion image dimension to the vgg input dimension, which is the default 244 input dimension.

…

On Fri, Dec 27, 2019 at 6:34 PM Alex J. Reyes ***@***.***> wrote: @zero-craft <https://github.com/zero-craft> by_name=True simply looks for layers with names of layers in the model, and tries to cast those same layers from the weights file. This circumvents the error but essentially initializes the model with random weights. Has anyone been able to get a solution to : You are trying to load a weight file containing 0 layers into a model with 16 layers. I'm facing the same problem. Just wondering if you've been able to find a fix or another weight file. Thanks! — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <https://gist.github.com/07d7802847aaad0a35d3?email_source=notifications&email_token=AOBV3GBGR2FFKK4LRKQJYJTQ2YKOJA5CNFSM4HHECR5KYY3PNVWWK3TUL52HS4DFVNDWS43UINXW23LFNZ2KUY3PNVWWK3TUL5UWJTQAF6RCI#gistcomment-3121700>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AOBV3GGEP3Q7RZGPOYHCDPDQ2YKOJANCNFSM4HHECR5A> .

Hi,
The input size of images is 2828. However I want to train them via VGG 16 where input size of image should be 244244.
How can I do that? I tried several methods but I am not successful.
I found flow_from_directory() function from keras but it requires path of directory, how I am directly accessing data from load_data().

Could you please tell me how can I achieve this?

F o r t h e l a z y :)

from keras import backend as K
from keras.models import Sequential
from keras.layers.core import Flatten, Dense, Dropout
from keras.layers.convolutional import Conv2D, MaxPooling2D, ZeroPadding2D
from keras.layers.core import Activation
from keras.optimizers import SGD
import cv2, numpy as np

K.set_image_data_format('channels_first')

def VGG_16(weights_path=None):
    model = Sequential()
    
  
    model.add(ZeroPadding2D((1,1),input_shape=(3,224,224)))
    model.add(Conv2D(64, (3, 3),padding='same'))
    model.add(Activation('relu'))  


    model.add(ZeroPadding2D((1,1)))
    model.add(Conv2D(64, (3, 3),padding='same'))
    model.add(Activation('relu'))     
    model.add(MaxPooling2D(pool_size=(2,2), strides=(2,2)))

    model.add(ZeroPadding2D((1,1)))
    model.add(Conv2D(128, (3, 3),padding='same'))
    model.add(Activation('relu')) 

    model.add(ZeroPadding2D((1,1)))
    model.add(Conv2D(128, (3, 3),padding='same'))
    model.add(Activation('relu')) 
    model.add(MaxPooling2D(pool_size=(2,2), strides=(2,2)))

    model.add(ZeroPadding2D((1,1)))
    model.add(Conv2D(256, (3, 3),padding='same'))
    model.add(Activation('relu')) 

    model.add(ZeroPadding2D((1,1)))
    model.add(Conv2D(256, (3, 3),padding='same'))
    model.add(Activation('relu')) 

    model.add(ZeroPadding2D((1,1)))
    model.add(Conv2D(256, (3, 3),padding='same'))
    model.add(Activation('relu')) 
    model.add(MaxPooling2D(pool_size=(2,2), strides=(2,2)))

    model.add(ZeroPadding2D((1,1)))
    model.add(Conv2D(512, (3, 3),padding='same'))
    model.add(Activation('relu')) 

    model.add(ZeroPadding2D((1,1)))
    model.add(Conv2D(512, (3, 3),padding='same'))
    model.add(Activation('relu')) 

    model.add(ZeroPadding2D((1,1)))
    model.add(Conv2D(512, (3, 3),padding='same'))
    model.add(Activation('relu')) 
    model.add(MaxPooling2D(pool_size=(2,2), strides=(2,2)))

    model.add(ZeroPadding2D((1,1)))
    model.add(Conv2D(512, (3, 3),padding='same'))
    model.add(Activation('relu')) 

    model.add(ZeroPadding2D((1,1)))
    model.add(Conv2D(512, (3, 3),padding='same'))
    model.add(Activation('relu')) 

    model.add(ZeroPadding2D((1,1)))
    model.add(Conv2D(512, (3, 3),padding='same'))
    model.add(Activation('relu')) 
    model.add(MaxPooling2D(pool_size=(2,2), strides=(2,2)))
    

    model.add(Flatten())
    model.add(Dense(4096))
    model.add(Activation('relu')) 
    model.add(Dropout(0.5))
    
    model.add(Dense(4096))
    model.add(Activation('relu')) 
    model.add(Dropout(0.5))

    model.add(Dense(1000))
    model.add(Activation('softmax')) 


    if weights_path:
        model.load_weights(weights_path,by_name=True)

    return model

if __name__ == "__main__":
    im = cv2.resize(cv2.imread('cat.jpg'), (224, 224)).astype(np.float32)
#     im[:,:,0] -= 103.939
#     im[:,:,1] -= 116.779
#     im[:,:,2] -= 123.68
    im = im.transpose((2,0,1))
    im = np.expand_dims(im, axis=0)
    

    model = VGG_16('vgg16_weights.h5')
    sgd = SGD(lr=0.1, decay=1e-6, momentum=0.9, nesterov=True)
    model.compile(optimizer=sgd, loss='categorical_crossentropy')
    out = model.predict(im)
    print (np.argmax(out))

Do you still have the weight file vgg16_weights.h5, I found the original google drive link is broken.

file is in user's trash

@baraldilorenzo (and @Zebreu!) thanks for this!

Is there a way I can use this to actually label images? I can get an index into the class vector (my 'cat.jpg' comes out to class 669) but I've no idea how to actually find the class labels the index refers to...

Tensorflow outputs in alphabetical order so if you have a list of all classes, sort them and find the label at index 669

The above-mentioned topic includes the long historical timeline discussions on the VGG16 model and need time to consume. I am thinking whether the following keras models can be used.

A Concise VGG16 Model
Release Date: Dec 3, 2019
https://github.com/1297rohit/VGG16-In-Keras

Official Keras VGG16 Model by the keras team
Release Data: Mar 29, 2019
https://github.com/keras-team/keras-applications/blob/master/keras_applications/vgg16.py

Can you please provide the weight files the above download link isn't working.

Thanks for your response. While using the weights provided by François Chollet, it can run. In this condition, I use Chollet's code with VGG16 model.

is the SGD the optimizer? I want to write the same to the paper in code, i didnt read the SGD, somebody friends.. i dont understand , this code is the same from the paper , even the optimizer? ..

@mikedewar You can download it at http://dl.caffe.berkeleyvision.org/caffe_ilsvrc12.tar.gz (there are other files too, but one of them has a thousand lines with synset IDs and names, so you can just use that as an index). Works well in my case.
Also, from the out array, you can do numpy.argmax(out) to get the class with the highest probability.

Also, there is no normalization done in the gist above. If you want accurate results, you better do those steps to any input image:

    img = cv2.resize(cv2.imread('../../Downloads/cat2.jpg'), (224, 224))

    mean_pixel = [103.939, 116.779, 123.68]
    img = img.astype(np.float32, copy=False)
    for c in range(3):
        img[:, :, c] = img[:, :, c] - mean_pixel[c]
    img = img.transpose((2,0,1))
    img = np.expand_dims(img, axis=0)

The mean pixel values are taken from the VGG authors, which are the values computed from the training dataset.

is the sgd the exactly optimizer from the paper ? , like the code .. or only the model vgg16 .. i want the exxactly code , in the paper dont mencione the OPTIMIZER

does anyone have a copy of the weights or any working link for download. the link provided seems dead at this moment

qu3ntinprime you can find the weights here: https://github.com/fchollet/deep-learning-models/releases.

I got this error when runing the code vgg16: "return tf.random.uniform ( ResourceExhaustedError: failed to allocate memory [Op:AddV2]". Could someone help me?

Unable to download vgg16.h5 file,Can please solve this issue.

Unable to download vgg16.h5 file,Can please solve this issue.

https://github.com/fchollet/deep-learning-models/releases/download/v0.1/vgg16_weights_tf_dim_ordering_tf_kernels.h5

这是来自QQ邮箱的假期自动回复邮件。   你好，你的邮件已收到,我会尽快给你回复。

i was able to download

…

On Mon, Dec 26, 2022 at 11:25 AM 张松伟 ***@***.***> wrote: ***@***.**** commented on this gist. ------------------------------ 这是来自QQ邮箱的假期自动回复邮件。 你好，你的邮件已收到,我会尽快给你回复。 — Reply to this email directly, view it on GitHub <https://gist.github.com/07d7802847aaad0a35d3#gistcomment-4413714> or unsubscribe <https://github.com/notifications/unsubscribe-auth/AC2YPX4MZ3EUVE6VK7PSOYTWPFXAXBFKMF2HI4TJMJ2XIZLTSKBKK5TBNR2WLJDHNFZXJJDOMFWWLK3UNBZGKYLEL52HS4DFQKSXMYLMOVS2I5DSOVS2I3TBNVS3W5DIOJSWCZC7OBQXE5DJMNUXAYLOORPWCY3UNF3GS5DZVRZXKYTKMVRXIX3UPFYGLK2HNFZXIQ3PNVWWK3TUUZ2G64DJMNZZDAVEOR4XAZNEM5UXG5FFOZQWY5LFVAZDMMJRGU2TEM5HORZGSZ3HMVZKMY3SMVQXIZI> . You are receiving this email because you commented on the thread. Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub> .

-- Grazie Eng. Gulli, Ph.D.