search

open-mmlab / mmdetection

OpenMMLab Detection Toolbox and Benchmark
https://mmdetection.readthedocs.io
Apache License 2.0
29.15k stars 9.39k forks source link

Labelling Training Set - Bounding Boxes #7823

Open optimaltaskin opened 2 years ago

optimaltaskin commented 2 years ago

Hi,

I have a question about training set labelling. Because I'm trying to detect an objects mask, I have been labelling my images as below:

Screenshot from 2022-04-25 08-29-55

I only selected mask boundaries for mask detection and didn't select any bounding boxes. Do I need to select bounding boxes as well for more proper results? Moreover when I don't select bounding boxes, how the algorithm compares its results with the test set since there is no bounding box in the set?

Czm369 commented 2 years ago

Maybe you can generate bboxes from masks, referring to https://github.com/open-mmlab/mmdetection/blob/c72bc707e661d61cf09aca0a53ad21812ef874d0/mmdet/core/mask/utils.py#L68

optimaltaskin commented 2 years ago

@Czm369 thanks for your response. However I wonder if it is required to draw bounding boxes as well? I mean does the result of training change when I use a set that only contains mask labels and another set that contains both mask and bounding box labels?