{+7}zfxfB.%RWQ;6.:c0f?k}

> Yes I have found the same problem. Luckily I have already converted OI to COCO format previously. Hi bro how you make through this? May you please show us...

> try using different images having dimensions of (512+1_64), (512+2_64), (512+3_64) ..... (512-1_64), (512-2_64), (512-3_64)... Uhh, sorry I don't quite get what the numbers mean..

> gradient_from_gpu_1 + gradient_from_gpu_2 Thanks @lllyasviel ! So basically bottleneck is the one holding gradient averaging, while remaining should work fine, _e.g._, GPU0 require 24G+24G; GPU1 require 24G; GPU2 require...

Sry I got question again. I travel from recognition community. In recognition, normally the multi-GPU training won't result in significant different RAMs among GPUs. Does this "1-big-gpu" thing only happen...

> go google search then type keyword is "link href" Great, thanks for reply ;D

@jangjichang Hi Jiangjichang, I'm working on the project too. This project uses keras1 and also have some libs that is out of date. I have also met the same problem...