Instance Segmentation on COCO (minival)
44.4AP^maskBoTNet-200
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| BoTNet-200Backbone=BoT200, Epochs=72, Multi-scale jitter=[0.1, 2.0]2021.01 | 44.4 | 68.9 | 48.2 | — | — | — | |
| BoTNet-152Backbone=BoT152, Epochs=72, Multi-scale jitter=[0.1, 2.0]2021.01 | 43.7 | 68.2 | 47.4 | — | — | — | |
| RevBiFPN-S6Backbone=RevBiFPN-S6, Params=130.2M, MACs=518.5B, Memory=7.4GB, Learning Schedule=1x, Framework=Mask R-CNN, Resolution=800x13332022.06 | 38.7 | — | — | 19.8 | 41.7 | 55.2 | |
| RevBiFPN-S5Backbone=RevBiFPN-S5, Params=80.5M, MACs=382.0B, Memory=5.5GB, Learning Schedule=1x, Framework=Mask R-CNN, Resolution=800x13332022.06 | 37.8 | — | — | 18.5 | 40.7 | 54.3 | |
| HRNETV2P-W32Backbone=HRNETV2P-W32, Params=49.9M, MACs=352.0B, Memory=9.0GB, Learning Schedule=2x, Framework=Mask R-CNN, Resolution=800x13332022.06 | 37.6 | — | — | 17.8 | 40 | 55 | |
| RevBiFPN-S4Backbone=RevBiFPN-S4, Params=55.5M, MACs=304.1B, Memory=4.1GB, Learning Schedule=1x, Framework=Mask R-CNN, Resolution=800x13332022.06 | 37.1 | — | — | 17.8 | 40.1 | 53.4 | |
| HRNETV2P-W32Backbone=HRNETV2P-W32, Params=49.9M, MACs=352.0B, Memory=9.0GB, Learning Schedule=1x, Framework=Mask R-CNN, Resolution=800x13332022.06 | 36.7 | — | — | 17.3 | 39 | 53 | |
| RESNET-101-FPNBackbone=RESNET-101-FPN, Params=63.2M, MACs=349.7B, Memory=5.8GB, Learning Schedule=2x, Framework=Mask R-CNN, Resolution=800x13332022.06 | 36.7 | — | — | 17 | 39.5 | 54.8 | |
| R101 GNBackbone=ResNet-101, Normalization=Group Normalization (GN), Training protocol=from scratch2018.03 | 36.4 | 58.2 | 38.7 | — | — | — | |
| RESNET-101-FPNBackbone=RESNET-101-FPN, Params=63.2M, MACs=349.7B, Memory=5.8GB, Learning Schedule=1x, Framework=Mask R-CNN, Resolution=800x13332022.06 | 36.1 | — | — | 16.2 | 39 | 53 | |
| RevBiFPN-S3Backbone=RevBiFPN-S3, Params=33.0M, MACs=232.9B, Memory=2.6GB, Learning Schedule=1x, Framework=Mask R-CNN, Resolution=800x13332022.06 | 35.5 | — | — | 17.4 | 38.4 | 50.9 | |
| HRNETV2P-W18Backbone=HRNETV2P-W18, Params=30.1M, MACs=249.3B, Memory=6.7GB, Learning Schedule=2x, Framework=Mask R-CNN, Resolution=800x13332022.06 | 35.3 | — | — | 16.9 | 37.5 | 51.8 | |
| R50 GNBackbone=ResNet-50, Normalization=Group Normalization (GN), Training protocol=from scratch2018.03 | 35.2 | 56.9 | 37.6 | — | — | — | |
| RESNET-50-FPNBackbone=RESNET-50-FPN, Params=44.2M, MACs=269.8B, Memory=4.2GB, Learning Schedule=2x, Framework=Mask R-CNN, Resolution=800x13332022.06 | 35 | — | — | 16 | 37.5 | 52 | |
| RESNET-50-FPNBackbone=RESNET-50-FPN, Params=44.2M, MACs=269.8B, Memory=4.2GB, Learning Schedule=1x, Framework=Mask R-CNN, Resolution=800x13332022.06 | 34.2 | — | — | 15.7 | 36.8 | 50.2 | |
| HRNETV2P-W18Backbone=HRNETV2P-W18, Params=30.1M, MACs=249.3B, Memory=6.7GB, Learning Schedule=1x, Framework=Mask R-CNN, Resolution=800x13332022.06 | 33.8 | — | — | 15.6 | 35.6 | 49.8 | |
| RevBiFPN-S2Backbone=RevBiFPN-S2, Params=26.5M, MACs=210.5B, Memory=2.6GB, Learning Schedule=1x, Framework=Mask R-CNN, Resolution=800x13332022.06 | 33.7 | — | — | 16 | 35.9 | 49.2 | |
| RevBiFPN-S1Backbone=RevBiFPN-S1, Params=23.1M, MACs=193.7B, Memory=2.4GB, Learning Schedule=1x, Framework=Mask R-CNN, Resolution=800x13332022.06 | 31 | — | — | 14.1 | 33.3 | 45.3 | |
| RevBiFPN-S0Backbone=RevBiFPN-S0, Params=22.2M, MACs=188.2B, Memory=2.0GB, Learning Schedule=1x, Framework=Mask R-CNN, Resolution=800x13332022.06 | 29.7 | — | — | 13.5 | 32.3 | 44.2 |