Skip to content

Conversation

@hellock
Copy link
Member

@hellock hellock commented May 10, 2019

Difference from the implementation of mmcv.runner.parallel_test:

  • The new implementation uses mmddp, and there can only be one process on each gpu, while parallel_test supports multiple processes per gpu.
  • The speed and gpu usage is much higher than parallel_test.

This PR is also related to #577. SyncBN cannot work with original parallel_test but should work well with mmddp.

@hellock hellock merged commit 2bd4c84 into open-mmlab:master May 11, 2019
@hellock hellock deleted the multigpu-test branch May 21, 2019 03:38
JegernOUTT pushed a commit to JegernOUTT/mmdetection that referenced this pull request Nov 23, 2019
* rewrite multigpu testing with mmddp

* fix an indent

* update slurm testing script

* update readme and scripts
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant