-
Notifications
You must be signed in to change notification settings - Fork 9.4k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
label value in CustomDataset #3629
Comments
I check Codebase Conventions,
So, for |
Yes. As you have quoted, in MMDetection 2.0 the background class is put at the end and class 0 is the first foreground class. |
According to https://github.com/open-mmlab/mmdetection/blob/master/mmdet/models/dense_heads/rpn_head.py Line 22, background_label=0. |
For CustomDataset, you label value is 0.
…On Sat, Sep 19, 2020, 22:09 Shujian Deng ***@***.***> wrote:
According to
https://github.com/open-mmlab/mmdetection/blob/master/mmdet/models/dense_heads/rpn_head.py
Line 22, background_label=0.
I'm quite confused now. When preparing my annotations, what value should I
assign for my foreground items? (assuming that I only have 1 foreground)
—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
<#3629 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AAWP34LEBFXPPWPH4ZOCN73SGS3R5ANCNFSM4QMPEFGA>
.
|
did u notice the code? it states that background_label=0. So I'm not sure if there is something wrong. |
@emergencyd Check https://mmdetection.readthedocs.io/en/latest/compatibility.html#codebase-conventions, for |
I know it is stated that we should use K for the background. But the code shows that 0 is for the background. |
@emergencyd you mean this line?
But, according to
use_sigmoid=true .
|
No. As I mentioned before, please check
and
@hubutui |
The |
Describe the bug
Hi, I'm new in object detection. I have some experience in classification and segmentation. We usually regard label value
0
as background. However, it seems mmdetection regards label value0
as the first foreground? I test with KittiTiny dataset, and here are my scripts.Reproduction
script to convert KittiTiny to mmdet's middle format
kitti_detection.py
and
classes.txt
faster_rcnn_r50_fpn_1x_kitti.py
tool/train.py
after dataset are build:Environment
python mmdet/utils/collect_env.py
to collect necessary environment information and paste it here.conda install -c pytorch pytorch cudatoolkit=10.1.243 torchvision
Outputs
So the label values are
0, 1, 2
for 3 classes,0
is not background.The text was updated successfully, but these errors were encountered: