joelthchao/tensorflow-finetune-flickr-style

Demonstrate the finetune process on TensorFlow

tensorflow-finetune-flickr-style's Issues

I reach 0.4056 in test set using fine tunning

2018-07-02 08:16:08.890713 Iter 12160: Testing Accuracy = 0.4056

Why you are only 0.3250? So big variance of training, what happened?

can't get dateset

can you share you dateset in baidu cloud, it seem hard to get the dataset because of network time out!

question for test.txt file

hi,Thanks for your share first. I want to know what is the test.txt file look like? In this test.txt file, if a label is need for every image?

ValueError

Hi，I am so sorry to trouble you again, but I have a issue troubled me these days. I use your code https://github.com/joelthchao/tensorflow-finetune-flickr-style/blob/master/finetune.py to finetune the network onto my own data to classify kaggle Dogs VS. Cats. It runs this error: ValueError: Cannot feed value of shape () for Tensor u'Placeholder:0', which has shape '(50, 227, 227, 3)'
Why did it happened? Please help me.

Problem for finetune VGG16 model

Hi, I try to modify your finetune.py and model.py in order to fine-tune VGG16 where converted from Caffe. I extracted VGG16_Tensorflow.npy via converting VGG16 Caffe model. The code shows error when I added "load_with_skip('VGG16_Tensorflow.npy', sess, ['fc8']) ", the error information is

Traceback (most recent call last):
  File "mff_vgg16.py", line 125, in <module>
    main()
  File "mff_vgg16.py", line 66, in main
    load_with_skip('VGG16_Tensorflow.npy', sess, ['fc8']) # Skip weights from fc8
  File "/home/larson/atensorflow_code/Document14C/finetune_example/network.py", line 19, in load_with_skip
    session.run(tf.get_variable(subkey).assign(data))
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/variables.py", line 497, in assign
    return state_ops.assign(self._variable, value, use_locking=use_locking)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/gen_state_ops.py", line 45, in assign
    use_locking=use_locking, name=name)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/op_def_library.py", line 455, in apply_op
    as_ref=input_arg.is_ref)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/ops.py", line 620, in convert_to_tensor
    ret = conversion_func(value, dtype=dtype, name=name, as_ref=as_ref)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/constant_op.py", line 179, in _constant_tensor_conversion_function
    return constant(v, dtype=dtype, name=name)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/constant_op.py", line 162, in constant
    tensor_util.make_tensor_proto(value, dtype=dtype, shape=shape))
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/tensor_util.py", line 353, in make_tensor_proto
    _AssertCompatible(values, dtype)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/tensor_util.py", line 290, in _AssertCompatible
    (dtype.name, repr(mismatch), type(mismatch).__name__))
TypeError: Expected float32, got 'weights' of type 'str' instead.

VGG16 model is

@staticmethod
    def VGG16(_X, _dropout, n_class):
        # part 1 (conv-conv-pool)
        conv1_1 = conv(_X, 3, 3, 64, 1, 1, padding='SAME', name='conv1_1')
        conv1_2 = conv(conv1_1, 3, 3, 64, 1, 1, padding='SAME', name='conv1_2')
        pool1 = max_pool(conv1_2, 2, 2, 2, 2, padding='SAME', name='pool1')
        # part 2 (conv-conv-pool)
        conv2_1 = conv(pool1, 3, 3, 128, 1, 1, padding='SAME', name='conv2_1')
        conv2_2 = conv(conv2_1, 3, 3, 128, 1, 1, padding='SAME', name='conv2_2')
        pool2 = max_pool(conv2_2, 2, 2, 2, 2, padding='SAME', name='pool2')
        # part 3 (conv-conv-conv-pool)
        conv3_1 = conv(pool2, 3, 3, 256, 1, 1, padding='SAME', name='conv3_1')
        conv3_2 = conv(conv3_1, 3, 3, 256, 1, 1, padding='SAME', name='conv3_2')
        conv3_3 = conv(conv3_2, 3, 3, 256, 1, 1, padding='SAME', name='conv3_3')
        pool3 = max_pool(conv3_3, 2, 2, 2, 2, padding='SAME', name='pool3')
        # part 4 (conv-conv-conv-pool)
        conv4_1 = conv(pool3, 3, 3, 512, 1, 1, padding='SAME', name='conv4_1')
        conv4_2 = conv(conv4_1, 3, 3, 512, 1, 1, padding='SAME', name='conv4_2')
        conv4_3 = conv(conv4_2, 3, 3, 512, 1, 1, padding='SAME', name='conv4_3')
        pool4 = max_pool(conv4_3, 2, 2, 2, 2, padding='SAME', name='pool4')
        # part 5 (conv-conv-conv-pool)
        conv5_1 = conv(pool4, 3, 3, 512, 1, 1, padding='SAME', name='conv5_1')
        conv5_2 = conv(conv5_1, 3, 3, 512, 1, 1, padding='SAME', name='conv5_2')
        conv5_3 = conv(conv5_2, 3, 3, 512, 1, 1, padding='SAME', name='conv5_3')
        pool5 = max_pool(conv5_3, 2, 2, 2, 2, padding='SAME', name='pool5')
        # Layer 6 (fc-relu-drop)
        fc6 = tf.reshape(pool5, [-1, 7*7*512])
        fc6 = fc(fc6, 7*7*512, 4096, name='fc6')
        fc6 = dropout(fc6, _dropout)
        # Layer 7 (fc-relu-drop)
        fc7 = fc(fc6, 4096, 4096, name='fc7')
        fc7 = dropout(fc7, _dropout)
        # Layer 8 (fc-prob)
        fc8 = fc(fc7, 4096, n_class, relu=False, name='fc8')
        return fc8

How to modify the code ?
Thanks a lot !

maybe overfitting?

• thanks for you code. now i am run the code .
• but i fond maybe the model overfitting.

2019-05-21 16:26:02.627367 Iter 5040: Training Loss = 0.0148, Accuracy = 1.0000
2019-05-21 16:26:07.453140 Iter 5060: Training Loss = 0.0249, Accuracy = 1.0000
2019-05-21 16:26:12.326071 Iter 5080: Training Loss = 0.0172, Accuracy = 1.0000
2019-05-21 16:26:17.227455 Iter 5100: Training Loss = 0.0552, Accuracy = 0.9800
2019-05-21 16:26:27.450521 Iter 5120: Testing Accuracy = 0.4336

• there we can see , train loss is lower and train accuracy is near 1.0 ;while the test accuracy is 0.4336 .
• later i will make some change to make test accuracy more high. I don't know if my understanding is right 。

How do you know the style mapping?

Hi, thanks for your remarkable work! I want to download photos with more than 20 styles in your code. I search a lot on flickr website, but cannot find any document about the style mapping. Can you give me a link which talk about this? Thanks!

How to save the model weights again!

How to download caffenet.npy?

Hello,when I use this project and I have read the readme ,But I still don't konw how to download the caffenet.npy.please I tell me how to download that,or you can give me a Internet connection。Thank you very much!

cannot gdownload any images

Downloading 80000 images with 3 workers...
Writing train/val for 0 successfully downloaded images.

SyntaxError: invalid syntax

When I wanna download the flickr dataset,an error occurs:
python assemble_data.py images train.txt test.txt 500
File "assemble_data.py", line 67
print('{} {}'.format(url, class_id), file=f)
^
SyntaxError: invalid syntax
Do you know what the reason is?

Is the code for finetuning?

Dear author, thanks for your code, I am working on it. but I find that maybe it is not for finetuning as the model is not pretrained.


how to load a pretrained model? would you like to paste the code?
another question, what're the .txt files look like? I want to make them with my own dataset as the some urls in .csv are invalid.
thanks again, any help appreiated!