Machine Learning

1 readers

1 users here now

Community Rules:

Be nice. No offensive behavior, insults or attacks: we encourage a diverse community in which members feel safe and have a voice.
Make your post clear and comprehensive: posts that lack insight or effort will be removed. (ex: questions which are easily googled)
Beginner or career related questions go elsewhere. This community is focused in discussion of research and new projects that advance the state-of-the-art.
Limit self-promotion. Comments and posts should be first and foremost about topics of interest to ML observers and practitioners. Limited self-promotion is tolerated, but the sub is not here as merely a source for free advertisement. Such posts will be removed at the discretion of the mods.

founded 1 year ago

MODERATORS

communick@academy.garden

[R] Correct way to calculate IOU, recall and precision for the image segmentation task (alien.top)

submitted 11 months ago by Meddhouib10@alien.top to c/machinelearning@academy.garden

4 comments fedilink hide all child comments

Hello,

I have a question that may seem trivial but which confused me because I found different approaches to doing it in different papers.

So let’s take the recall as an example : to calculate it on the test set I found two different methods in the codes available for published papers. Some calculate the recall for every image then calculate the mean on all images, other simply use the formula as if each pixel was an element in it’s own. What is the correct way to doing so ?

Thanks in advance !

top 4 comments

sorted by: hot top controversial new old

[–] VAL9THOU@alien.top 1 points 11 months ago

As long as the specific method they use is documented (and, ideally, their ground truth and results) I would say it doesn't really matter. There aren't really any single correct methods for scoring most ML models, AFAIK

[–] jeandebleau@alien.top 1 points 11 months ago (1 children)

If all the images have the same size, it does not matter because in this case the sum of the means equals the mean of the sum.

[–] Meddhouib10@alien.top 1 points 11 months ago

How is that ? You devide by the either the pred positive pixel or the ground truth positive pixel depending of either you are calculating the recall or the precision

[–] puppet_pals@alien.top 1 points 11 months ago

Wrote https://arxiv.org/abs/2207.12120 but never published. I think the "metric definition" is a pretty nice explanation of how it works. Process is the same between bounding boxes & segmentation maps aside from the IoU computation.