Anyone know the input/output format for these auto-encoder onnx?

[elephantpanda]

I am running these autoencoder onnx using Onnxruntime in c#:

I am trying to encode/decode an image of a cat just using the vae-encoder and decoder.

I am getting a blurry outcome of the reconstituted cat.

I wonder if anyone knows the correct format for the input and output tensors?

I began by encoding my image as RGB with float32 values in the range [-1,1]
And the output of the vae-decoder I was running through a sigmoid then multiplying by 255 to get pixel values.

I have a feeling this is not correct. Does anyone know the correct format for the input/output tensors?

I could even be in the wrong colour-space for example maybe it wants HSL?

[xadupre]

Did you try in the range [0, 1]?