Below is a gray-scale image of a __  which is labeled as 'Original' of dimensions 32*32 pixels (total of 1024 pixels). The structure of Auto-Encoder used for compression here is [1024-256-64-8-64-256-1024], which means that a 1024 Dimension image is being compressed to just 8 Dimensions. And in the second part of the image it can be seen how well the 'Auto-Encoder' is able to reconstruct the 'Original' image. Just beside that, we can see how badly PCA algorithm reconstructs the same image after compressing the original image to 8 Dimensions (i.e. taking first 8 principal components only).
PS: Auto-Encoder compression and PCA are NOT loss-less compression techniques