EfficientAD batch size

[alexerupakat]

Why can't we give a higher batch size for training efficientAD? Why it can only train on batch size =1

[abc-125]

Because it is the parameter used in the paper.

[alexerupakat]

Thanks. But can you help me understand why the paper prefers batch size one?

[rishabh-akridata]

@alexerupakat Did you understand why they are using a batch size of 1 in EfficientAD?

[alexriedel1]

There is evidence that autoencoders might produce sharper output images if trained on smaller batch sizes, maybe it's because of this.