Adds different types of relevance to the accuracy layer#3532
Open
timmeinhardt wants to merge 3 commits intoBVLC:masterfrom
Open
Adds different types of relevance to the accuracy layer#3532timmeinhardt wants to merge 3 commits intoBVLC:masterfrom
timmeinhardt wants to merge 3 commits intoBVLC:masterfrom
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This PR adds a type option to the accuracy layer, which can be set to either recall (REC), precision (PRE) or jaccard index (JAC). The default option is REC which corresponds to the previously existing implementation and therefore does not break any existing usages. All other options and the accuracy per class top blob work seamlessly with the different types.
Especially the jaccard index, which corresponds to the intersection over union, is a useful relevance measurement for segmentation tasks.
I want to point out that the layer name "accuracy" might be confused with the "rand accuracy" or "rand index" (sometimes only called "accuracy") which is a relevance measurement in itself. For the sake of consistency the layer should actually be called "relevance", but I wanted to keep backwards compatibility.
Also one should be aware of what using a top_k > 1 in combination with the jaccard index actually means. The accuracy per class will not always increase, because the intersection is effectively increased.
I appreciate any comments or improvements and hope that this feature is regarded as useful.