-
Notifications
You must be signed in to change notification settings - Fork 2.5k
Conversation
Thank you for your pull request and welcome to our community. We require contributors to sign our Contributor License Agreement, and we don't seem to have you on file. In order for us to review and merge your code, please sign up at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need the corporate CLA signed. If you have received this in error or have any questions, please contact us at cla@fb.com. Thanks! |
Thank you for signing our Contributor License Agreement. We can now accept your code for this (and any) Facebook open source project. Thanks! |
hi @fallingdust |
hi @zimenglan-sysu-512 ,
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This looks very reasonable to me.
One drawback is that I think it will be very slow, as it will compute a lot of mask predictions, which will be filtered afterwards. But as a first implementation this is great!
One thing I'd do differently is that I would not use a global config in the code, but let's not be blocked by changing this.
Thanks a lot!
import torch | ||
import torchvision.transforms as TT | ||
|
||
from maskrcnn_benchmark.config import cfg |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
using the config globally in the library is something I've tried to avoid, but let's merge this as is and maybe modify this later on
Since you disabled the bbox filtering after box head, you will have 1000 boxes fed into the mask head. How could you fit 1000 masks into GPU memory? |
I guess you test on faster rcnn, but the config file is mask rcnn. |
@fallingdust @shikunyu8 @fmassa |
Hi, @youngwanLEE |
* Implement multi-scale testing(bbox aug) like Detectron. * Add comment. * Fix missing cfg after merge.
Implemented bbox augmentations(h-flip, multi-scale, multi-scale h-flip) in test time, just like Detectron.
There's a little hack in the data loading process, if you have better solution, please tell me.