Instance Segmentation on MS COCO (test-dev)
38.3mAP@[.5:.95]MS R-CNN
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| MS R-CNNBackbone=R-101-FPN2019.04 | 38.3 | 58.8 | 41.5 | 17.8 | 40.4 | 54.4 | 8.6 | 116.3 | — | — | |
| MS R-CNNBackbone=R-101-FPN, FPS=8.6, Time=116.32019.12 | 38.3 | 58.8 | 41.5 | 17.8 | 40.4 | 54.4 | — | — | — | — | |
| FCISmulti-scale testing=true, horizontal flip=true, multi-scale training=true, ensemble=true, number of networks=62016.11 | 37.6 | 59.9 | — | — | — | — | — | — | — | — | |
| Mask R-CNNbackbone=ResNeXt-101-FPN2017.03 | 37.1 | 60 | 39.4 | 16.9 | 39.9 | 53.5 | — | — | — | — | |
| PA-NetBackbone=R-50-FPN2019.04 | 36.6 | 58 | 39.3 | 16.3 | 38.1 | 53.1 | 4.7 | 212.8 | — | — | |
| PA-NetBackbone=R-50-FPN, FPS=4.7, Time=212.82019.12 | 36.6 | 58 | 39.3 | 16.3 | 38.1 | 53.1 | — | — | — | — | |
| Mask R-CNNbackbone=ResNet-101-FPN2017.03 | 35.7 | 58 | 37.8 | 15.5 | 38.1 | 52.4 | — | — | — | — | |
| Mask R-CNNBackbone=R-101-FPN2019.04 | 35.7 | 58 | 37.8 | 15.5 | 38.1 | 52.4 | 8.6 | 116.3 | — | — | |
| Mask R-CNNBackbone=R-101-FPN, FPS=8.6, Time=116.32019.12 | 35.7 | 58 | 37.8 | 15.5 | 38.1 | 52.4 | — | — | — | — | |
| Mask R-CNNsupervision=Full2021.03 | 35.7 | 58 | 37.8 | — | — | — | — | — | — | — | |
| RetinaMaskBackbone=R-101-FPN2019.04 | 34.7 | 55.4 | 36.9 | 14.3 | 36.7 | 50.5 | 6 | 166.7 | — | — | |
| RetinaMaskBackbone=R-101-FPN, FPS=6, Time=166.72019.12 | 34.7 | 55.4 | 36.9 | 14.3 | 36.7 | 50.5 | — | — | — | — | |
| YOLACT-550++Backbone=R-101-FPN, FPS=27.3, Time=36.72019.12 | 34.6 | 53.8 | 36.9 | 11.9 | 36.8 | 55.1 | — | — | — | — | |
| Mask R-CNNBackbone=ResNet-50, DA Policy=DADA2020.03 | 34.5 | — | — | — | — | — | — | — | — | — | |
| Mask R-CNNBackbone=ResNet-50, DA Policy=baseline2020.03 | 34.2 | — | — | — | — | — | — | — | — | — | |
| YOLACT-550++Backbone=R-50-FPN, FPS=33.5, Time=29.92019.12 | 34.1 | 53.3 | 36.2 | 11.7 | 36.1 | 53.6 | — | — | — | — | |
| G-RMI (2016)2016.11 | 33.8 | 56.9 | — | — | — | — | — | — | — | — | |
| FCISmulti-scale testing=true, horizontal flip=true, multi-scale training=true, ensemble=false2016.11 | 33.6 | 54.5 | — | — | — | — | — | — | — | — | |
| FCIS+++ + OHEMbackbone=ResNet-101-C5-dilated, OHEM=true, multi-scale train/test=true, horizontal flip test=true2017.03 | 33.6 | 54.5 | — | — | — | — | — | — | — | — | |
| Mask R-CNNbackbone=ResNet-101-C42017.03 | 33.1 | 54.9 | 34.8 | 12.1 | 35.6 | 51.1 | — | — | — | — | |
| FCISmulti-scale testing=true, horizontal flip=true, multi-scale training=false, ensemble=false2016.11 | 32.7 | 52.7 | — | — | — | — | — | — | — | — | |
| FCISmulti-scale testing=true, horizontal flip=false, multi-scale training=false, ensemble=false2016.11 | 32 | 51.9 | — | — | — | — | — | — | — | — | |
| YOLACT-700Backbone=R-101-FPN, Input Size=7002019.04 | 31.2 | 50.6 | 32.8 | 12.1 | 33.3 | 47.1 | 23.4 | 42.7 | — | — | |
| YOLACT-700Backbone=R-101-FPN, FPS=23.4, Time=42.72019.12 | 31.2 | 50.6 | 32.8 | 12.1 | 33.3 | 47.1 | — | — | — | — | |
| YOLACT-550Backbone=R-101-FPN, Input Size=5502019.04 | 29.8 | 48.5 | 31.2 | 9.9 | 31.3 | 47.7 | 33.5 | 29.8 | — | — | |
| YOLACT-550Backbone=R-101-FPN, FPS=33.5, Time=29.82019.12 | 29.8 | 48.5 | 31.2 | 9.9 | 31.3 | 47.7 | — | — | — | — | |
| FCISBackbone=R-101-C52019.04 | 29.5 | 51.5 | 30.2 | 8 | 31 | 49.7 | 6.6 | 151.5 | — | — | |
| FCISBackbone=R-101-C5, FPS=6.6, Time=151.52019.12 | 29.5 | 51.5 | 30.2 | 8 | 31 | 49.7 | — | — | — | — | |
| FCIS baselinemulti-scale testing=false, horizontal flip=false, multi-scale training=false, ensemble=false2016.11 | 29.2 | 49.5 | — | — | — | — | — | — | — | — | |
| FCIS + OHEMbackbone=ResNet-101-C5-dilated, OHEM=true2017.03 | 29.2 | 49.5 | — | 7.1 | 31.3 | 50 | — | — | — | — | |
| YOLACT-550Backbone=D-53-FPN, Input Size=5502019.04 | 28.7 | 46.8 | 30 | 9.5 | 29.6 | 45.5 | 40.7 | 24.6 | — | — | |
| YOLACT-550Backbone=D-53-FPN, FPS=40.7, Time=24.62019.12 | 28.7 | 46.8 | 30 | 9.5 | 29.6 | 45.5 | — | — | — | — | |
| MNC+++ (2015)2016.11 | 28.4 | 51.6 | — | — | — | — | — | — | — | — | |
| YOLACT-550Backbone=R-50-FPN, Input Size=5502019.04 | 28.2 | 46.6 | 29.2 | 9.2 | 29.3 | 44.8 | 45 | 22.2 | — | — | |
| YOLACT-550Backbone=R-50-FPN, FPS=45, Time=22.22019.12 | 28.2 | 46.6 | 29.2 | 9.2 | 29.3 | 44.8 | — | — | — | — | |
| BBAMsupervision=Box2021.03 | 25.7 | 50 | 23.3 | — | — | — | — | — | — | — | |
| FAIRCNN (2015)2016.11 | 25 | 45.6 | — | — | — | — | — | — | — | — | |
| YOLACT-400Backbone=R-101-FPN, Input Size=4002019.04 | 24.9 | 42 | 25.4 | 5 | 25.3 | 45 | 45.3 | 22.1 | — | — | |
| YOLACT-400Backbone=R-101-FPN, FPS=45.3, Time=22.12019.12 | 24.9 | 42 | 25.4 | 5 | 25.3 | 45 | — | — | — | — | |
| Multi-task Network Cascades (MNC)backbone=ResNet-1012015.12 | 24.6 | 44.3 | — | — | — | — | — | — | — | — | |
| MNCbackbone=ResNet-1012015.12 | 24.6 | 44.3 | — | — | — | — | — | — | — | — | |
| MNCbackbone=ResNet-101-C42017.03 | 24.6 | 44.3 | 24.8 | 4.7 | 25.9 | 43.6 | — | — | — | — | |
| Multi-task Network Cascades (MNC)backbone=VGG-162015.12 | 19.5 | 39.7 | — | — | — | — | — | — | — | — | |
| MNCbackbone=VGG-162015.12 | 19.5 | 39.7 | — | — | — | — | — | — | — | — | |
| LIIDsupervision=Image label2021.03 | 16 | 27.1 | 16.5 | — | — | — | — | — | — | — | |
| Fan et al.supervision=Image label, Instance saliency2021.03 | 13.7 | 25.5 | 13.5 | — | — | — | — | — | — | — | |
| Cascade-RCNNBackbone=ResNeXt101, Deformable=n/a, Multi-scale inference=false2020.04 | — | — | — | — | — | — | — | — | 45.8 | 38.6 | |
| Cascade-RCNNBackbone=ResNeSt101, Deformable=v2, Multi-scale inference=false2020.04 | — | — | — | — | — | — | — | — | 50 | 43 | |
| DCNv2Backbone=ResNet101, Deformable=v2, Multi-scale inference=false2020.04 | — | — | — | — | — | — | — | — | 45.8 | 39.7 | |
| DCNv2Backbone=ResNeXt101, Deformable=v2, Multi-scale inference=false2020.04 | — | — | — | — | — | — | — | — | 46.7 | 40.5 | |
| SNIPERBackbone=ResNet101, Deformable=v1, Multi-scale inference=true2020.04 | — | — | — | — | — | — | — | — | 47.1 | 41.3 |