about one GPU corresponds to multiple images

[foralliance]

@bharatsingh430 HI

Section 5 of the paper points out:
Simple modifications like 5 scales for RPN anchors, a minimum image size of 800, 16 images per minibatch and 256 ROIs per image were 2. The Training was done on 8 GPUs in parallel.

For 16 images per minibatch, 8 GPUs, equivalent to 2 pictures per the GPU
In rfcn_edn2end_ohen_coco. yml, IMS_PER_BATCH = 1.
In solver_ohem. prototxt, iter_size = 2,

Is it as long as the product of IMS_PER_BATCH and iter_size satisfies 2，it equivalent to a GPU corresponding to the two images.
Is that correct?

---

Because the existing py-r-fcn and py-faster RCNN only support IMS_PER_BATCH=1. Therefore, if a GPU corresponds to multiple images, can it only be satisfied by adjusting the parameter iter_size?
That's my understanding, I'm not sure it is rights.

So appreciated for your reply.