search

michaelgroeger / boxshrink

This is the repository for the corresponding 2022 MICCAI-MILLanD workshop paper "BoxShrink: From Bounding Boxes to Segmentation Masks"
MIT License
22 stars 1 forks source link

License and question #2

Open Jpvmello opened 1 year ago

Jpvmello commented 1 year ago

Hi, nice work! I would like to ask two questions:

  1. I saw no information about the repo's license. Is it okay to use/modify, even for a commercial project?
  2. Did you evaluate the method on images with overlapping bounding boxes? I mean: if the foreground inside a box contain part of the foreground inside another box, how would the system deal with it?
michaelgroeger commented 1 year ago

Hi @Jpvmello,

Thanks for reaching out and the interest in our work!

1) I added an MIT License so you can use it in commercial projects. It would just be nice if you cite our work if possible.

2) I am not 100 % sure if there are images like that in the dataset. I skimmed through the first 120 and what comes closest to your question are images 70-77. The system should be able to deal with an overlap though. It would treat it as one area and then the objects might be separated after comparing the embeddings and/or the CRF stage. Note however that we don't do instance segmentation so the mask(s) will be of the same class.

If you have further questions don't hesitate to ask.

Best,

Michael