No account yet ?

Upsampling

One of the ways to upsample the compressed image is to decluster using nearest neighbor or maximum decluster.

Another way is to use transpose convolution.

The convolution operation with strides results in undersampling.

The transpose convolution is the inverse of the convolution operation.

Here the kernel is placed on the pixels of the input image.

The pixel values are successively multiplied by the kernel weights to produce the oversampled image.

If there is an overlap, the values are added together.

Kernel weights in oversampling are learned in the same way as in convolutional operation, which is why it is also called learnable oversampling.

Another way is to use the oversampling and nearest neighbor convolution layers in Decoder instead of transposing the convolution layers.

This method prevents checkerboard artifacts in images caused by transpose convolution.

Source : https://iq.opengenus.org/autoencoder/