Out of memory

[njtuzzy]

@kuangliu I am using a titian X graphic card with 12G memory , somehow, when i use the bach_size=2, there are still out or memory, may i ask the configuration of hardware and software you are using? thanks

[kuangliu]

This is a big model. I use batch_size=16 with 8 P40 (24G memory each). You can modify the fpn, like use RetinaFPN18:

def RetinaFPN18():
    return RetinaFPN(Bottleneck, [2,2,2,2])

And also decrease the input_size from 600 to like 400. But I don't think you can squeeze it into one 12G memory GPU...

[njtuzzy]

Acctually, when i downsample the input image size to 224, it will take about 4G memory size and should be ok for training, the disavantage is that the down scale is a little bit large :(

thanks

[acgtyrant]

You have 8 P40, so crazy.