ovalery16 / swap-face Goto Github PK

Hello

I'm porting your work over to Colaboratory, but having trouble with the preprocessing function during face extraction.
For some reason, I keep getting the error during the preprocessing process, and the loop terminates prematurely.

Some images do get preprocessed, so I'm not sure what's the problem. Using the Daniel Craig images in your data folder works fine.

"Failed to extract from image: ../content/chris/chris (80).jpg. Reason: list index out of range"

Code below, any advice would be helpful!

EDIT: So I took a closer look, it's definitely certain images that causes system trip. I believe it's due to the face_recognition library, name faces_filter.py. If no face is detected, and hence no encoding, then the resulting array will be null, and hence an exception is thrown. Will update here in due course

`

Hi, when I tried to run the train.ipynb it went wrong! So I edit it in pycharm, it shows this. What can I do to train on new data? Thanks! @OValery16

Hi

Ive managed to produce some outputs the results are quite blurry, so I've been trying out MSE instead of MAE. However, I've noticed the weights you given us are from the bond transfer, do you happen to have the original weights for trump to cage or whatnot?

Thank you

Regards

Adrian

Hi ,

I tried face-swapping between Nickolas cage and Donald Trump Dataset but my output of the test script contains a blur grayscale patch over the face detected in trumps faces in the images of my test data. The scripts seem to run and finish without throwing up any errors and I trained the model without any specific error. Can someone help me with the possible reason be behind this happening?
I have uploaded 2/5 output images after performing the execution of prediction scripts
The images used are of size 256X256 while training and testing and I just used the same images( 5 images) for from training data to perform the test.
I used around 375 images for trump faces training and approx 310 for cage training faces. I am stuck at this point and can't proceed further to reach any conclusion

Tha

AttributeError: module 'keras.utils.conv_utils' has no attribute 'normalize_data_format'

Got this error due to unsupported version of Keros and Tensorflow

My version in system is Keros=2.2.2 and Tensorflow=1.13.0

Can you tell me the correct version of both of these

Hello, don't know if you are the original author of this tutorial – if so, thank you for making this so easy to understand for beginners!

Hope you can help me understand the Autoencoder model better:

IMAGE_SHAPE = (64, 64, 3)
ENCODER_DIM = 1024

    def Encoder(self):
        input_ = Input(shape=IMAGE_SHAPE)
        x = input_
        x = self.conv(128)(x)
        x = self.conv(256)(x)
        x = self.conv(512)(x)
        x = self.conv(1024)(x)
        x = Dense(ENCODER_DIM)(Flatten()(x))
        x = Dense(4 * 4 * 1024)(x)
        x = Reshape((4, 4, 1024))(x)
        x = self.upscale(512)(x)
        return KerasModel(input_, x)

    def Decoder(self):
        input_ = Input(shape=(8, 8, 512))
        x = input_
        x = self.upscale(256)(x)
        x = self.upscale(128)(x)
        x = self.upscale(64)(x)
        x = Conv2D(3, kernel_size=5, padding='same', activation='sigmoid')(x)
        return KerasModel(input_, x)

What would be an efficient way to increase the input image size (128, 192, 256) so that the Decoder is still producing good results? Which values are important – which should be increased? The filters of the Encoder or Decoder? Simply more convolution layers? Or is the ENCODER_DIM more important?

Why does the Decoder have this specific input shape? Is it connected to the input image size?
input_ = Input(shape=(8, 8, 512))