Train and test on different size datasets

[saumitrabg]

We are looking at YOLACT++ inference of 4096x756 image at a high FPS. However, to not run out of memory, our train/val dataset is 756x756. We are seeing low scores when we are running inference on 4096x756. I see the issue is still open: #319

Do you have any guidance here?

[PrajwalCogniac]

Hey @saumitrabg
Did you get any interesting finds on this issue ? , I am facing the same problem as well