Dice coefficient

A mask-overlap metric that weights shared area twice against the sum of both masks' areas.

The Dice coefficient scores overlap between a predicted mask $A$ and ground truth $B$ as $\text{Dice}=2\lvert A\cap B\rvert/(\lvert A\rvert+\lvert B\rvert)$ -- related to IoU by $\text{Dice}=2\,\text{IoU}/(1+\text{IoU})$ , always at least as large as IoU for the same pair of masks, and the more commonly reported metric for segmentation masks specifically.

Examples

A predicted mask overlapping ground truth with shared area $30$ out of total areas $50$ and $40$ scores $\text{Dice}=2(30)/(50+40)\approx0.667$ , noticeably higher than that same pair's IoU of $0.5$ .