A problom about batch size

[scnuhealthy]

When I train the model with coco dataset and fpn-cascade model, it seems that the batch size can only be 1. How can I train the model with larger batch size?

[zzdang]

I don't know where to set batch size. Could you tell me?

[ackbar03]

I see that it can be set in the train.prototext and test.prototext.

I also set batch size to 1 for training because of memory issue, however after every 2500 iterations the solver runs it once over the test set and I run out of memory again.

I already set both train and test batch size to 1 though. Does anyone have any idea if there there is anything else I need to set? Or is it running out of memory because it needs to hold both the data from the training and the test set at the same time, in which case it seems that I will be unable to run this model?

[ackbar03]

I see that it can be set in the train.prototext and test.prototext.

I also set batch size to 1 for training because of memory issue, however after every 2500 iterations the solver runs it once over the test set and I run out of memory again.

I already set both train and test batch size to 1 though. Does anyone have any idea if there there is anything else I need to set? Or is it running out of memory because it needs to hold both the data from the training and the test set at the same time, in which case it seems that I will be unable to run this model?