List your company
ResearchFormerly WizWand

Panoptic Segmentation on COCO (test-dev)

56.2PQPanoptic SegFormer
40.28844.41948.5552.681Jan 8, 2019Aug 5, 2019Mar 1, 2020Sep 27, 2020Apr 24, 2021Nov 19, 2021Jun 17, 2022
Updated 8mo ago

Evaluation results

MethodLinks
Panoptic SegFormerBackbone=Swin-L2021.0956.282.867.162.3——47——
Panoptic SegFormerBackbone=PVTv2-B5, Pre-trained on ImageNet-22K=true2021.0955.88366.561.9——46.5——
CMT-DeepLabbackbone=Axial-R104-RFN, params=270.3M, pre-training=ImageNet-22K, training iterations=200k2022.0655.783.665.961.6——46.8——
CMT-DeepLabbackbone=Axial-R104-RFN, params=270.3M, pre-training=ImageNet-22K2022.0655.483.565.661——47——
K-NetBackbone=Swin-L, Pre-trained on ImageNet-22K=true2021.0955.282.466.161.2——46.2——
K-Netbackbone=Swin-L, pre-training=ImageNet-22K2022.0655.282.466.161.2——46.2——
CMT-DeepLabbackbone=Axial-R104, params=135.2M, pre-training=ImageNet-22K2022.0654.583.264.759.6——46.9——
InnovationBackbone=ensemble2021.0953.583.463.361.8——41.1——
CMT-DeepLabbackbone=Axial-R50, params=94.9M, pre-training=ImageNet-22K2022.0653.48363.658.3——46——
MaskFormerBackbone=Swin-L, Pre-trained on ImageNet-22K=true2021.0953.38264.159.1——44.5——
MaskFormerbackbone=Swin-L, params=212M, pre-training=ImageNet-22K2022.0653.38264.159.1——44.5——
Megvii (Face++)backbone=ensemble model2019.0153.283.262.962.285.572.539.579.748.5
MaX-DeepLab-LBackbone=MaX-L, TTA=false2020.1251.3——57.2——42.4——
Max-Deeplab-LBackbone=Max-L2021.0951.382.561.357.2——42.4——
MaX-DeepLab-Lbackbone=MaX-L, params=451M2022.0651.382.561.357.2——42.4——
Panoptic SegFormerBackbone=R1012021.0950.98261.256.2——43——
Panoptic SegFormerBackbone=R502021.0950.281.960.455.3——42.4——
DetectoRSTTA=true2020.0650——58.5——37.2——
DetectoRSBackbone=ResNeXt-101, MS=true2020.1149.6——57.8——37.1——
DetectoRSBackbone=RX-101, TTA=true2020.1249.6——57.8——37.1——
MaX-DeepLab-SBackbone=MaX-S, TTA=false2020.1249——54——41.6——
MaX-DeepLab-Sbackbone=MaX-S, params=61.9M2022.0649——54——41.6——
Ada-Segmentbackbone=ResNet-101-FPN-D2020.1248.581.858.255.7——37.6——
SOGNetBackbone=DCN-101, Multi-scale=true2020.0347.8————————
SOGNetTTA=true2020.0647.8————————
SOGNetbackbone=ResNet-101-FPN-D2020.1247.880.757.6——————
SOGNetBackbone=DCN-101, MS=true2020.1147.8————————
Panoptic FCNParadigm=box-free, Backbone=DCN101-FPN, Simple enhanced version=true, val set for training=true2020.1247.5——53.7——38.2——
SpatialFlowBackbone=DCN-101, Multi-scale=true2020.0347.3——53.5——37.9——
SpatialFlowTTA=true2020.0647.3——53.5——37.9——
SpatialFlowbackbone=ResNet-101-FPN-D2020.1247.381.856.953.5——37.9——
BANetbackbone=ResNet-101-FPN-D2020.1247.380.857.554.9——35.9——
SpatialFlowBackbone=DCN-101, MS=true2020.1147.3——53.5——37.9——
BANetParadigm=box-based, Backbone=DCN101-FPN2020.1247.3——54.9——35.9——
Li et al.Backbone=DCN-101, Multi-scale=true2020.0347.2——53.5——37.7——
Li et al.TTA=false2020.0647.2——53.5——37.7——
Li et al.Backbone=DCN-101, MS=true2020.1147.2——53.5——37.7——
Unifying+Paradigm=box-based, Backbone=DCN101-FPN2020.1247.2——53.5——37.7——
Panoptic FCNParadigm=box-free, Backbone=DCN101-FPN, Simple enhanced version=true2020.1247.1——53.2——37.8——
Panoptic FCNParadigm=box-free, Backbone=DCN101-FPN2020.1247——53——37.8——
Caribbeanbackbone=ensemble model2019.0146.880.557.154.381.865.935.578.543.8
OCFusionBackbone=ResNeXt-101, Multi-scale testing=true, Deformable convolution=true2019.0646.7——54——35.7——
OCFusionbackbone=ResNeXt-101-FPN-D2020.1246.7——54——35.7——
Panoptic-DeepLabBackbone=SWideRNet-(1, 1, 4), MS=true2020.1146.7——52.2——38.3——
UPSNetbackbone=ResNet-101-FPN2019.0146.680.556.953.281.564.636.778.945.3
UPSNet-msBackbone=ResNet-101-FPN, Multi-scale test=true2019.0946.6——53.2——36.7——
UPSNetBackbone=DCN-101, Flip=true, MS=true2019.1146.6——53.2——36.7——
UPSNetBackbone=ResNet-101, Multi-scale testing=true, Deformable convolution=true2019.0646.6——53.2——36.7——
UPSNetBackbone=DCN-101, Multi-scale=true2020.0346.6——53.2——36.7——
UPSNetTTA=true2020.0646.6——53.2——36.7——
UPSNetbackbone=ResNet-101-FPN-D2020.1246.680.556.953.2——36.7——
UPSNetBackbone=DCN-101, MS=true2020.1146.6——53.2——36.7——
UPSNetParadigm=box-based, Backbone=DCN101-FPN2020.1246.6——53.2——36.7——
UPSNetStage=Two Stage, Backbone=ResNet-101-FPN, Flip=true, MS=true2020.0346.6——53.2——36.7——
UPSNetBackbone=DCN-101, TTA=true2020.1246.6——53.2——36.7——
AUNetbackbone=ResNeXt-152-FPN2019.0146.58156.155.983.766.332.57740.7
AUNetBackbone=ResNeXt-152-FPN2019.0946.5——55.9——32.5——
AUNetBackbone=ResNeXt-152, Flip=false, MS=false2019.1146.5——55.8——32.5——
AUNetBackbone=ResNeXt-152, Multi-scale testing=true, Deformable convolution=false2019.0646.5——55.9——32.5——
AUNetBackbone=ResNeXt-152, Multi-scale=false2020.0346.5——55.8——32.5——
AUNetTTA=false2020.0646.5——55.8——32.5——
AUNetbackbone=ResNeXt-152-FPN-D2020.1246.58156.155.8——32.5——
AUNetBackbone=ResNeXt-152, MS=false2020.1146.5——55.8——32.5——
AUNetParadigm=box-based, Backbone=ResNeXt152-FPN2020.1246.5——55.8——32.5——
AUNetStage=Two Stage, Backbone=ResNeXt-101-FPN, Flip=true, MS=false2020.0346.5——55.8——32.5——
PKU_360backbone=ResNeXt-152-FPN2019.0146.379.656.158.683.769.627.673.635.6
DETRBackbone=ResNet-101, MS=false2020.1146————————
DETRBackbone=RN-101, TTA=false2020.1246————————
Panoptic FCNParadigm=box-free, Backbone=Res101-FPN2020.1245.5——51.4——36.4——
CIAEParadigm=box-based, Backbone=DCN101-FPN2020.1244.5——49.7——36.8——
Axial-DeepLab-LBackbone=Axial-ResNet-L, Multi-scale=true2020.0344.2——49.2——36.8——
Axial-DeepLab-LTTA=true2020.0644.2——49.2——36.8——
Axial-DeepLab-LBackbone=Axial-ResNet-L, MS=true2020.1144.2——49.2——36.8——
Axial-DeepLab-LBackbone=AX-L, TTA=true2020.1244.2——49.2——36.8——
Axial-DeepLab-LBackbone=Axial-ResNet-L, Multi-scale=false2020.0343.6——48.9——35.6——
Axial-DeepLabParadigm=box-free, Backbone=Axial-ResNet-L2020.1243.6——48.9——35.6——
Axial-DeepLab-LBackbone=AX-L, TTA=false2020.1243.6——48.9——35.6——
Axial-DeepLab-MBackbone=Axial-ResNet-M, Multi-scale=false2020.0343.2——48.1——35.9——
Panoptic FPNBackbone=LVT, Params (M)=5.4, FLOPs (G)=56.4, FPS (1333, 800)=20.42021.1243——49.9——32.6——
AdaptISBackbone=ResNeXt-1012019.0942.8——50.1——31.8——
AdaptISBackbone=ResNeXt-101, Flip=false, MS=false2019.1142.8——53.2——36.7——
AdaptISBackbone=ResNeXt-101, Multi-scale testing=false, Deformable convolution=false2019.0642.8——50.1——31.8——
AdaptISBackbone=ResNeXt-101, Multi-scale=true2020.0342.8——53.2——36.7——
AdaptISTTA=true2020.0642.8——53.2——36.7——
AdaptISBackbone=ResNeXt-101, MS=true2020.1142.8——53.2——36.7——
AdaptISParadigm=box-free, Backbone=ResNeXt-1012020.1242.8——53.2——36.7——
AdaptISStage=Two Stage, Backbone=ResNeXt-101, Flip=false, MS=false2020.0342.8——50.1——31.8——
Axial-DeepLab-SBackbone=Axial-ResNet-S, Multi-scale=false2020.0342.2——46.5——35.7——
Panoptic-DeepLabBackbone=Xception-71, Flip=true, MS=true2019.1141.4——45.1——35.9——
Panoptic-DeepLabBackbone=Xception-71, Multi-scale=true2020.0341.4——45.1——35.9——
Panoptic-DeepLabTTA=true2020.0641.4——45.1——35.9——
Panoptic-DeepLabBackbone=Xception-71, MS=false2020.1141.4——45.1——35.9——
Panoptic DeepLabBackbone=X-71, TTA=true2020.1241.4——45.1——35.9——
OANetBackbone=ResNet-101, Multi-scale testing=false, Deformable convolution=false2019.0641.3——50.4——27.7——
OANetbackbone=ResNet-101-FPN2020.1241.3——50.4——27.7——
Panoptic FPNBackbone=PVTv2-B0, Params (M)=5.3, FLOPs (G)=49.7, FPS (1333, 800)=23.52021.1241.2——47.7——31.5——
Panoptic FPNbackbone=ResNet-101, model_type=single network2019.0140.9——48.3——29.7——
Panoptic-FPNBackbone=ResNet-101, Flip=false, MS=false2019.1140.9——48.3——29.7——
Panoptic FPNBackbone=ResNet-101, Multi-scale testing=false, Deformable convolution=false2019.0640.9——48.3——29.7——
Panoptic-FPNBackbone=ResNet-101, Multi-scale=false2020.0340.9——48.3——29.7——