Skip to content

Made separate functions for computing output shapes for all layers. R…#1156

Merged
opencv-pushbot merged 1 commit intoopencv:masterfrom
arrybn:layers_shapes
May 23, 2017
Merged

Made separate functions for computing output shapes for all layers. R…#1156
opencv-pushbot merged 1 commit intoopencv:masterfrom
arrybn:layers_shapes

Conversation

@arrybn
Copy link
Copy Markdown
Contributor

@arrybn arrybn commented May 11, 2017

…emoved output blobs allocation from layers

This pullrequest changes

Added functions for computing layers shapes without inference. Moved creating blobs from layers to Net

@arrybn arrybn force-pushed the layers_shapes branch 4 times, most recently from 5652f12 to 6c3ba83 Compare May 11, 2017 13:14
@arrybn
Copy link
Copy Markdown
Contributor Author

arrybn commented May 11, 2017

@vpisarev please take a look. I've changed Net interface by adding new functions

@vpisarev
Copy link
Copy Markdown
Contributor

vpisarev commented May 17, 2017

getOutShapes() has been introduced, but it's not used in the corresponding allocate() methods. Also, as we discussed, getOutShapes() should also return size of a temporary buffer. And also it should report whether inputs can be processed in-place.

Well, actually, allocate() method should be modified. They should not allocate memory either (and thus they should be renamed). In dnn2 branch I renamed allocate() to finalize() and made it superset of the proposed getOutShapes() method. I think, something like that should be done. The memory should be allocated by the Net engine

@arrybn
Copy link
Copy Markdown
Contributor Author

arrybn commented May 23, 2017

@vpisarev I've made all changes, which you've mentioned in your comment above. Please take a look

@vpisarev
Copy link
Copy Markdown
Contributor

vpisarev commented May 23, 2017

👍

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.

4 participants