ERROR:torch.distributed.elastic.multiprocessing.api:failed (exitcode: -9) local_rank: 0 (pid: 71517) about detr3d HOT 7 OPEN

自问自答：换个大内存的机器再试一下吧，比如40GB的

from detr3d.

Hi,
You need around 20 GB to fit ResNet101 with BatchSize as 1.
For VovNet, you need larger.
See similar: #21

from detr3d.

Hi, do you have solved it? I have the same problem as you.

from detr3d.

Hi, do you have solved it? I have the same problem as you.

He seems to have memory issue, and he fixed this after using a GPU with higher memory.

You need around 20GB to train ResNet-101 with FPN with batch-size as 1 on a single GPU. To train VoVNet, you need higher

Do you see any cuda out of memory in your error log?

from detr3d.

Hi, do you have solved it? I have the same problem as you.

He seems to have memory issue, and he fixed this after using a GPU with higher memory.

You need around 20GB to train ResNet-101 with FPN with batch-size as 1 on a single GPU. To train VoVNet, you need higher

Do you see any cuda out of memory in your error log?

I do see 'cuda out of memory' in my error log. How can i do to solve this problem?
Thank you.

from detr3d.

Hi, do you have solved it? I have the same problem as you.

He seems to have memory issue, and he fixed this after using a GPU with higher memory.
You need around 20GB to train ResNet-101 with FPN with batch-size as 1 on a single GPU. To train VoVNet, you need higher
Do you see any cuda out of memory in your error log?

I do see 'cuda out of memory' in my error log. How can i do to solve this problem? Thank you.

Maybe you can reduce the batchsize.

from detr3d.

it's not useful

Hi, do you have solved it? I have the same problem as you.

He seems to have memory issue, and he fixed this after using a GPU with higher memory.
You need around 20GB to train ResNet-101 with FPN with batch-size as 1 on a single GPU. To train VoVNet, you need higher
Do you see any cuda out of memory in your error log?

I do see 'cuda out of memory' in my error log. How can i do to solve this problem? Thank you.

Maybe you can reduce the batchsize.

but i use this method without any benifit

from detr3d.