[
https://issues.apache.org/jira/browse/MXNET-40?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16392408#comment-16392408
]
ASF GitHub Bot commented on MXNET-40:
-------------------------------------
wkcn commented on issue #9939: [MXNET-40]add multi proposal operator (cpu
version) and fix the bug in proposal op (gpu version)
URL: https://github.com/apache/incubator-mxnet/pull/9939#issuecomment-370285916
@pengzhao-intel @xinyu-intel
Thank you! I will have a try.
The performance table
name|time (ms)
------|----------
BBoxTransformInv|268
IoUTransformInv|Not used
FilterBox|22
CopyScore|18
ReverseArgsort(unstable sort)|7303
ReorderProposals|338
nms(calculate area)|286
nms(calcuate nms)|7547
allocate memory for workspace|1
copy anchor to workspace_proposal|0
enumrate all shifted anchors|9
copy workspace_proposals_base to workspace_proposals|162
assign foreground scores for each anchor|45
prepare output|3
Total|16002
Using stable sort to sort anchors (ReverseArgsort) will increase about 3000
ms.
Update: 2018-03-09 (626296b)
name|time (ms)
------|----------
BBoxTransformInv|523
FilterBox|35
CopyScore|49
ReverseArgsort(unstable sort)|1036
ReorderProposals|17
nms|2573
allocate memory for workspace|4
GenerateAnchors|0
copy anchor and score to workspace_proposal|361
enumerate all shifted anchors|101
prepare output|0
Total|4699
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
> add multi proposal operator (cpu version) and fix the bug in proposal op (gpu
> version)
> --------------------------------------------------------------------------------------
>
> Key: MXNET-40
> URL: https://issues.apache.org/jira/browse/MXNET-40
> Project: Apache MXNet
> Issue Type: Improvement
> Reporter: Kan Wu
> Priority: Major
>
> The multi_proposal operator ({{mxnet.sym.contrib.MultiProposal}}, CPU
> version) is not implemented before.
> I wrote the code about it.
> And I found there was a bug in _proposal.cu_ and _multi_proposal.cu_.
> The {{batch_size}} of the output of _Proposal_ and _MultiProposal_ are both
> {{param_.rpn_post_nms_top_n}}.
> When {{count_anchors < param_.rpn_post_nms_top_n}}, the variable
> {{rpn_post_nms_top_n}} will be {{count_anchors}}, which is less than
> {{param_.rpn_post_nms_top_n}}.
> [https://github.com/apache/incubator-mxnet/blob/master/src/operator/contrib/proposal.cu#L438]
> It will cause the problem that the output whose index is larger than
> {{rpn_post_nms_top_n}} will be _not assigned_.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
