-
Notifications
You must be signed in to change notification settings - Fork 306
Conversation
ef13378
to
2eb19d9
Compare
I guess the advantage of |
Yea |
I made a mix-in for this problem. Please check #454. |
Areas of instance masks are needed for evaluation. So, I added an option that makes the dataset to return them. I've been working on evaluation of COCO as well. My work can be found here. |
@yuyu2172 Could you merge master branch? |
ok |
|
||
There are total of 82,783 training and 40,504 validation images. | ||
'minval' split is a subset of validation images that constitutes | ||
5000 images in the validation images. The remaining validation |
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.
5,000 images
?
docs/source/reference/datasets.rst
Outdated
@@ -51,6 +51,13 @@ CityscapesTestImageDataset | |||
~~~~~~~~~~~~~~~~~~~~~~~~~~ | |||
.. autoclass:: CityscapesTestImageDataset | |||
|
|||
COCO |
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.
I prefer MS COCO
because we use PASCAL VOC
in this doc.
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.
LGTM
Depends on #452
I think we need some modification to the BboxDataset API.
(0,4)
.labels
for all bbox datasets. The method returns a list of bounding boxes. Also, I suggest to support similar method for all label datasets.