Table 2.
Instance segmentation mask on COCO (Common Objects in Context) test-dev. FCIS was the winner of COCO 2016. FCIS+++ uses more mature technology. Mask R-CNN’s network segmentation accuracy has been substantially improved, and it is the benchmark for the 2017 segmentation network model. PAN, proposed in 2018, yielded excellent results; however, the network is complex, the amount of data is large, and it requires more time. MS R-CNN in 2019 yielded the best results. MR R-CNN is suitable for predicting large and medium-sized objects, while other indicators are close to the best value.
Method | Backbone | AP | AP50 | AP75 | APS | APM | APL |
---|---|---|---|---|---|---|---|
FCIS | ResNet-101 | 29.2 | 49.5 | - | 7.1 | 31.3 | 50.0 |
FCIS+++ | ResNet-101 | 33.6 | 54.5 | - | - | - | - |
Mask R-CNN | ResNet-101-C4 | 33.1 | 54.9 | 34.8 | 12.1 | 35.6 | 51.1 |
Mask R-CNN | ResNet-101 FPN | 35.7 | 58.0 | 37.8 | 15.5 | 38.1 | 52.4 |
Mask R-CNN | ResNeXt-101 FPN | 37.1 | 60.0 | 39.4 | 16.9 | 39.9 | 53.5 |
PAN | ResNet-50 FPN | 38.2 | 60.2 | 41.4 | 19.1 | 41.1 | 52.6 |
MS R-CNN | ResNet-101 | 35.4 | 54.9 | 38.1 | 13.7 | 37.6 | 53.3 |
MS R-CNN | ResNet-101 FPN | 38.3 | 58.8 | 41.5 | 17.8 | 40.4 | 54.4 |
MS R-CNN | ResNet-101 DCN-FPN | 39.6 | 60.7 | 43.1 | 18.8 | 41.5 | 56.2 |
MR R-CNN | ResNet-50 FPN | 35.2 | 53.5 | 39.8 | 13.9 | 38.1 | 52.6 |
MR R-CNN | ResNet-101 FPN | 37.6 | 56.1 | 41.1 | 16.4 | 40.6 | 54.7 |
MR R-CNN | ResNet-101 DCN-FPN | 38.8 | 58.0 | 42.7 | 17.2 | 41.8 | 56.6 |