Hi,Contributor，I have a question and would like to invite you to discuss it together. Are the metrics you proposed suitable for the two modes shown in the image?
example 1：

exexam 2：

Looking forward to your reply

Hi Bowen,

I really like that you made a metric which focuses on the boundary quality rather than overall IOU. I was applying your method on my models and was not sure exactly how to use the ground truth boxes for evaluation (Section 6.2, Table 4). I have the intuition that instead of taking the region predicted by RPN, you use the ground truth boxes' region and apply a ROIAlign operation to get the region of interest. Am I correct?

Also, can you suggest a clean way of doing this?

Thanks in advance.

Best,
Hamd ul Moqeet Riaz

Hi Bowen,
Thanks for your excellent work.
I am newer about segmentation tasks. I would like to ask could you point the Mask IoU you are using in the paper to me?
Thank you!

After I install the boundary-iou-api : pip install git+https://github.com/bowenc0221/boundary-iou-api.git
I can not use the mmdetecion code : ./tools/./tools/dist_tra.sh

the erro is :
torch.distributed.elastic.multiprocessing.errors.ChildFailedError:

./tools/train.py FAILED

Hello
How are you?
Thanks for contributing to this project.
I have a question.
Can this be used as boundary loss in the segmentation model training?
Thanks

Originally posted by @rose-jinyang in #2 (comment)

Hi Author,
Thanks for your excellent work. I would like to ask whether the ground truth and prediction are binary masks (only includes 0 or 1)?
If the prediction and ground truth are not 0s or 1s, should we use a threshold value to make it to be a binary mask? If yes, how do we choose the threshold value, is it 128 or 0.5? If yes, why did we choose them as the threshold values? Thank you!

Dear Phd:
How can I use boundary-iou in my own datasets?

Hi,
I am trying to evaluate my dataset using masks and boundaries. I am seeing that the AP values are exactly same. My instances are approximately 60x60 pixels. Any insights ?

Congratulations for the awesome work!

I have a doubt which is - How will the boundaries be generated if the objects are very small ? I observed, that it does not generate boundaries for let say a very small square mask of size 2 x 2.

Thanks!

I found your work very interesting and thinking about using the Boundary IoU metric to evaluate my segmentation models. However, I can only see instance and panoptic segmentation scripts in this repo.
Does the boundary IoU metric not applicable to pixel-level segmentation?