[mlpack] Fwd: Essential Deep Learning Modules in GSoC '19

[Rajiv Vaidyanathan]

Hi Marcus,

I was also concerned about completion on time but just wanted to give a
list of ideas. Among those, I think implementing YOLO and it's necessary
layers along with tests and documentation and ROI pooling should be doable
in this summer. Implementing them would also make the process of building
RCNNs easy in the future. Please let me know your thoughts. Please let me
know if anything more can be added to it.

Thanks and regards,
Rajiv
ᐧ

On Thu, 14 Mar 2019 at 17:46, Marcus Edel <marcus.edel at fu-berlin.de> wrote:

> Hello Rajiv,
>
> I like the idea, but perhaps we should focus on either 1 or 2,
> implementing both
> might be difficult to get done in time. Let me know what you think.
>
> Thanks,
> Marcus
>
> On 14. Mar 2019, at 09:39, Rajiv Vaidyanathan <
> rajiv.vaidyanathan4 at gmail.com> wrote:
>
> Hi Marcus,
>
> I did some more research about the Graph Neural Networks. Since I'm new to
> GNNs, I am not sure about how long it will take to implement all the
> necessary layers. Hence, I thought I'll work on something which I'm already
> familiar about, which is convolutional neural networks. I'm extremely
> interested in working on methods used for object detection. Object
> detection algorithms such as R-CNN, YOLO, etc. have become very popular due
> to their speed and accuracy. I feel that it would be a great addition to
> the MLPack library.
>
> I want to implement the following networks along with tests and
> documentation:
> 1. R-CNN and it's variants such as fast RCNN, Faster RCNN and Mask-R-CNN
> 2. YOLO
>
> For their implementation, the following have to be implemented:
> 1. ROI pooling
> 2. Region Proposal Network
> 3. Techniques: non-max suppression, intersection over union and anchor
> boxes
>
> Please let me know what you think about this. If you are fine with this
> idea, I'll do more research and make a brief proposal as to what needs be
> precisely done with respect to the MLPack code along with a rough timeline.
>
> Thanks and regards,
> Rajiv
> ᐧ
>
> On Sat, 2 Mar 2019 at 03:23, Marcus Edel <marcus.edel at fu-berlin.de> wrote:
>
>> Hello Rajiv,
>>
>> thanks again for the contributions. Implementing Graph Neural Networks is
>> quite
>> a challenge especially timewise but if you are up for that; we should
>> make sure
>> the timeline is reasonable including the milestones. Everything needs to
>> be
>> tests and stable at the end of the summer which often takes a lot of
>> time, but
>> as I said if you are up for the challenge this could be an interesting
>> project
>> for sure.
>>
>> Let us know what you think.
>>
>> Thanks,
>> Marcus
>>
>> On 28. Feb 2019, at 19:14, Rajiv Vaidyanathan <
>> rajiv.vaidyanathan4 at gmail.com> wrote:
>>
>> Dear Marcus, Mikhail and Shikhar,
>>
>> I am N Rajiv Vaidyanathan and I'm interested in the topic "Essential Deep
>> Learning Modules" in GSoC 19.
>>
>> Over the past month, I'm trying to get an understanding of the MLPack
>> codebase by making contributions. As of now, I have implemented SPSA
>> optimiser, Dice Loss function and currently working on Dense blocks.
>>
>> I recently read a paper on Graph Neural Networks and found it to be
>> fascinating as it works very well on non-Euclidian spaces such as social
>> networks and 3D images. I am interested in working on implementing this
>> network along with tests and documentation. I'm also interested in
>> improving the overall documentation of ann.
>>
>> Please let me know what you think :)
>>
>> Thanks and regards,
>> Rajiv
>> ᐧ
>> _______________________________________________
>> mlpack mailing list
>> mlpack at lists.mlpack.org
>> http://knife.lugatgt.org/cgi-bin/mailman/listinfo/mlpack
>>
>>
>> ᐧ
> _______________________________________________
> mlpack mailing list
> mlpack at lists.mlpack.org
> http://knife.lugatgt.org/cgi-bin/mailman/listinfo/mlpack
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://knife.lugatgt.org/pipermail/mlpack/attachments/20190314/a46bf203/attachment.html>