[Colloquium] REMINDER: 5/14 TTIC Colloquium: Ross Girshick, Facebook

[Mary Marre via Colloquium]

 *When:  *   Monday, May 14th at *10:30 am*

*Where: *   TTIC, 6045 S Kenwood Avenue, 5th Floor, Room 526

*Who: *       Ross Girshick,  Facebook


*Title:       *Scaling Up Visual Recognition

*Abstract:*
I will present two efforts to scale visual recognition systems far beyond
the regime that is typically studied today. First, we'll look at a new
empirical study in which we trained large ConvNets (up to 20x more mult-add
ops than ResNet-101) on datasets that are 1000x larger than ImageNet, but
which are only labeled with social media hashtags. In spite of the noisy
labels, the trained models show excellent transfer learning results. The
results also raise questions about how far we can reach with the current
trend of larger models and more data, suggesting that more fundamental
breakthroughs are still required.

In the second part of my talk, we'll look at expanding the slice of the
visual world that object detectors can see. I'll present recent work on a
method for training Mask R-CNN to produce instance segmentations for
thousands of object categories even when mask annotations are only
available for the 80 COCO dataset categories. I'll conclude by discussing a
nascent effort to build a large-scale instance segmentation dataset that
will enable rigorous benchmarking of detectors that can recognize thousands
of object categories.



Host: Michael Maire <mmaire at ttic.edu>



For more information on the colloquium series or to subscribe to the
mailing list,please see http://www.ttic.edu/colloquium.php




Mary C. Marre
Administrative Assistant
*Toyota Technological Institute*
*6045 S. Kenwood Avenue*
*Room 504*
*Chicago, IL  60637*
*p:(773) 834-1757*
*f: (773) 357-6970*
*mmarre at ttic.edu <mmarre at ttic.edu>*

On Sun, May 13, 2018 at 9:18 PM, Mary Marre <mmarre at ttic.edu> wrote:

> *When:  *   Monday, May 14th at *10:30 am*
>
> *Where: *   TTIC, 6045 S Kenwood Avenue, 5th Floor, Room 526
>
> *Who: *       Ross Girshick,  Facebook
>
>
> *Title:       *Scaling Up Visual Recognition
>
> *Abstract:*
> I will present two efforts to scale visual recognition systems far beyond
> the regime that is typically studied today. First, we'll look at a new
> empirical study in which we trained large ConvNets (up to 20x more mult-add
> ops than ResNet-101) on datasets that are 1000x larger than ImageNet, but
> which are only labeled with social media hashtags. In spite of the noisy
> labels, the trained models show excellent transfer learning results. The
> results also raise questions about how far we can reach with the current
> trend of larger models and more data, suggesting that more fundamental
> breakthroughs are still required.
>
> In the second part of my talk, we'll look at expanding the slice of the
> visual world that object detectors can see. I'll present recent work on a
> method for training Mask R-CNN to produce instance segmentations for
> thousands of object categories even when mask annotations are only
> available for the 80 COCO dataset categories. I'll conclude by discussing a
> nascent effort to build a large-scale instance segmentation dataset that
> will enable rigorous benchmarking of detectors that can recognize thousands
> of object categories.
>
>
>
> Host: Michael Maire <mmaire at ttic.edu>
>
>
>
> For more information on the colloquium series or to subscribe to the
> mailing list,please see http://www.ttic.edu/colloquium.php
>
>
>
>
> Mary C. Marre
> Administrative Assistant
> *Toyota Technological Institute*
> *6045 S. Kenwood Avenue*
> *Room 504*
> *Chicago, IL  60637*
> *p:(773) 834-1757*
> *f: (773) 357-6970*
> *mmarre at ttic.edu <mmarre at ttic.edu>*
>
> On Fri, May 11, 2018 at 5:20 PM, Mary Marre <mmarre at ttic.edu> wrote:
>
>> *When:  *   Monday, May 14th at *10:30 am*
>>
>> *Where: *   TTIC, 6045 S Kenwood Avenue, 5th Floor, Room 526
>>
>> *Who: *       Ross Girshick,  Facebook
>>
>>
>> *Title:       *Scaling Up Visual Recognition
>>
>> *Abstract:*
>> I will present two efforts to scale visual recognition systems far beyond
>> the regime that is typically studied today. First, we'll look at a new
>> empirical study in which we trained large ConvNets (up to 20x more mult-add
>> ops than ResNet-101) on datasets that are 1000x larger than ImageNet, but
>> which are only labeled with social media hashtags. In spite of the noisy
>> labels, the trained models show excellent transfer learning results. The
>> results also raise questions about how far we can reach with the current
>> trend of larger models and more data, suggesting that more fundamental
>> breakthroughs are still required.
>>
>> In the second part of my talk, we'll look at expanding the slice of the
>> visual world that object detectors can see. I'll present recent work on a
>> method for training Mask R-CNN to produce instance segmentations for
>> thousands of object categories even when mask annotations are only
>> available for the 80 COCO dataset categories. I'll conclude by discussing a
>> nascent effort to build a large-scale instance segmentation dataset that
>> will enable rigorous benchmarking of detectors that can recognize thousands
>> of object categories.
>>
>>
>>
>> Host: Michael Maire <mmaire at ttic.edu>
>>
>>
>>
>> For more information on the colloquium series or to subscribe to the
>> mailing list,please see http://www.ttic.edu/colloquium.php
>>
>>
>>
>>
>> Mary C. Marre
>> Administrative Assistant
>> *Toyota Technological Institute*
>> *6045 S. Kenwood Avenue*
>> *Room 504*
>> *Chicago, IL  60637*
>> *p:(773) 834-1757*
>> *f: (773) 357-6970*
>> *mmarre at ttic.edu <mmarre at ttic.edu>*
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20180514/4d053e14/attachment-0001.html>