Embedding space [D]

Hello everyone, I’m relatively new to this area of machine learning and currently experimenting with Variational Autoencoders (VAEs) to build an embedding space for an image dataset with images have different spatial dimensions, I cannot easily standardize them to a fixed size. My current approach uses adaptive pooling in the encoder to produce a fixed-dimensional latent representation, so the model can in principle handle variable input sizes. However, now the results are quite poor so far, and the learned