Encounted OOM while trying to train the stage1 model

[stevenk12]

I use the script 'uni_dist_train.sh' and try to train the stage1(track and map) model, but I encounter the error of 'OUT OF MEMORY'.
I use four 3090 GPUs, the total memory is 96GB. Recording to your words in 'Eval and Train', the first stage needs about 50GB memory. So what's the problem? Can u help me. Thanks!