Yann Henon
Yann Henon
Does it work if you keep the line `retinanet = torch.nn.DataParallel(retinanet).cuda()` and pass `CUDA_VISIBLE_DEVICES=0`? It should fall back to the correct behaviour
this is with `retinanet = torch.nn.DataParallel(retinanet).cuda()` ?
@luisriera I believe this is fixed by upgrading to python 3.6 @rvandeghen that seems correct, thanks for your input. If you have time to submit a PR to fix this...
Unfortunately like many personal projects, I just don't have the time needed to dedicate to this project. If there are people interested in being added as contributors so that they...
I use python3, torch 1.4.0, torchvision 0.5.0
Hi, any news on this front? I understand these matters can take time.
I'm not sure exactly what the underlying reason is. I know others have the same issue when trying to train retinanet on pascal voc. I suspect retinanet works less well...
Unfortunately not much can be done about training speed - more GPUs and/or better GPUs help. Would be interested to hear about performance with voc07+12, although I've heard from others...
This repo needs an update for the new pytorch version, I have not had time yet. On Tue, Nov 6, 2018, 12:25 PM TinaLiuCV Hi, > When I run the...