Organize work on new features Caffe

[Bartosz Ludwiczuk]

Hi,
last month I spend learning and using Caffe framework. And it work greate, all examples work for me. Caffe has a lot of usefull features, thanks for that.
And I want to start discussion about organizing work at Caffe Framework.
As we all know, Deep Learning is changing dramaticly fast. And many of contibutors want to send their code with new usefull features. But most of them are do not merge till now.
Example:
DenseNet #308
Early stopping #76

Sliding Window, Varying input/output size and Dense, multiscale extraction #189

 

In fact, these one could be really useful for Caffe community.

What I want to say, that there is a little mess for begginer. I do not know which features of Caffe are now available, which features are in progress (because as we can see, Open PR does not mean "in progress").

Maybe some-where should be created list of:

• working features
• in progress features (with link to appropiate repo)
• features which have PR

It could be sth like RoadMap and status ( I know that there us "Milestone" at github, but this is different from RoadMap)

Other thing is metting Caffe group. As I understand, there are some metting of Caffe group at Berkeley. Maybe you could relese some notes after each metting? This will give overview where Caffe is going at next step.

What do you think about it?

[Evan Shelhamer]

Hey Bartosz,

It's true that as the Caffe development and community grows it is important to clearly signal features, work in progress, and plans. More incremental use of the GitHub milestones and a periodic "Roasting Notes" on Caffe to the mailing list could help this.

Let's pick up this conversation after 11/14 once Berkeley has come out of deadline mode. Please do post any further thoughts on workflow and handling PR progress.

Evan Shelhamer

--
You received this message because you are subscribed to the Google Groups "Caffe Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to caffe-users...@googlegroups.com.
To post to this group, send email to caffe...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/caffe-users/fb06b91a-b4e4-4bad-bbd4-0a43ed2c4360%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Stefano Fabri]

Yes I completely agree that some organization policy is needed. We had more contributors and PR but without a clear review policy and a fixed merging team:

https://openhub.net/p/caffe

[Vitaly Lavrukhin]

Hi Evan,

Caffe is a great framework for Deep Learning. Thank you and all developers making it better!

Though right now it looks like there is no clear direction or roadmap showing what should be done next.

Well, actually, I've found recently slides of Yangqing Jia's CVPR2015 presentation 'Improving Caffe: Some Refactoring' [http://tutorial.caffe.berkeleyvision.org/caffe-cvpr15-improving.pdf]. It has nice ideas inspired by functional programming approach.

But again it's unclear whether it's approved by core developers refactoring plan. And if yes, then when is it going to be implemented.

Could you please comment on this?

--

Kind regards,

Vitaly