zzyykk725

你好，请问怎么设置多卡分布式训练，我设置的DDP跑起来LOSS值非常大，几千上万了，这正常吗？

1

> > > > 请问怎么开多卡训练呢？我设置了 os.environ['CUDA_VISIBLE_DEVICES'] = '4,5,6,7' 和 __C.CONST.DEVICE = '4,5,6,7' 运行脚本python train_pcn.py 一直没反应 > > > > > > > > > > > > 你好，请问你这问题解决了吗，我无论怎么设置这个os.environ['CUDA_VISIBLE_DEVICES']都没用，都会默认分布，而且还是不均匀的 > >...

多卡训练问题

1

213 38831.4209 12157.8579 12144.3472 12137.6324 38831.4209 214 3236.2828 1069.4618 1043.5928 1085.6686 3236.2828 215 727.4463 268.9736 260.6899 240.1202 727.4463 216 29291.5916 8654.7327 8651.6600 8631.5527 29291.5916 217 14858.0999 4987.3714 4951.6268 4980.7844 14858.0999...

MVP训练的训练路径

2

__C.DATASETS = edict() __C.DATASETS.COMPLETION3D = edict() __C.DATASETS.COMPLETION3D.CATEGORY_FILE_PATH = './datasets/Completion3D.json' __C.DATASETS.COMPLETION3D.PARTIAL_POINTS_PATH = '/path/to/datasets/Completion3D/%s/partial/%s/%s.h5' __C.DATASETS.COMPLETION3D.COMPLETE_POINTS_PATH = '/path/to/datasets/Completion3D/%s/gt/%s/%s.h5' __C.DATASETS.SHAPENET = edict() __C.DATASETS.SHAPENET.CATEGORY_FILE_PATH = './datasets/ShapeNet.json' __C.DATASETS.SHAPENET.N_RENDERINGS = 8 __C.DATASETS.SHAPENET.N_POINTS = 2048 __C.DATASETS.SHAPENET.PARTIAL_POINTS_PATH = '../data/PCN/%s/partial/%s/%s/%02d.pcd'...