Inference time cost vs batch size

[deepconv]

Question

Hi Author, fantastic work! My question is that when I try inference with the Swin-L model using batch size > 1, I observe the inference time linearly scales with the batch size. Can you shed some lights on why the batch inference does not reduce the average time cost per image?

Additional

No response