How to train SSD：Single Shot MultiBox Detector

[he mingde]

How to train  SSD：Single Shot MultiBox Detector fork from https://github.com/weiliu89/caffe/tree/ssd

when use the original train.prototxt ,the error happen:

AssertionError: Unsupported data layer type AnnotatedData

[Mohammad Moradi]

[jaydee]

I believe SSD uses a custom datum layer (AnnotatedData) vs the regular Caffe Datum which, as the error message indicates, isn't supported by DIGITS

[Jon Watte]

Presumably you can re-build the Caffe that DIGITS uses to include this layer type?

Sincerely,

jw

Sincerely,

Jon Watte


--
"I find that the harder I work, the more luck I seem to have." -- Thomas Jefferson

--
You received this message because you are subscribed to the Google Groups "DIGITS Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to digits-users+unsubscribe@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/digits-users/8ae2ff77-f59a-4c9b-b5f7-d7e0a5b39ff3%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

[Mohammad Moradi]

Hi
I rebuild Caffe with AnnotatedData layer and linked it to DIGITS but I got the same error. I think DIGITS should have supports for SDD since it is one of the state of art networks for object detection.

On Tuesday, April 18, 2017 at 1:23:18 PM UTC+4:30, he mingde wrote:

[Balint Magyar]

Hi Mohammad,

did you find any solution?

[huih...@gmail.com]

did you find a way to train SSD?

在 2017年4月18日星期二 UTC+8下午4:53:18，he mingde写道：