Does the glow’s bundle compiler support input with varying size? Like varying image size input?
I think this depends on the model. E.g. I’ve seen C2 models where the shapes of the network are inferred based on the input size, so you have to specify that at least when compiling. But other models may have specific shape expectations in the model and so aren’t flexible, you’d hit an error if you try different image sizes that are not compatible.