List your company

Instance Segmentation on MS-COCO (Standard AP Metrics)

50.8APDINAT-L
-1.51212.06925.6539.231Sep 29, 2022Apr 27, 2023Nov 23, 2023Jun 20, 2024Jan 16, 2025Aug 14, 2025Mar 13, 2026

Evaluation results

MethodLinks
DINAT-LFramework=Mask2Former [5], Win. Size=11 x 11, # of Params=220 M, FLOPs=522 G, Pre-trained=ImageNet-22K, Resolution=800x8002022.0950.87530.954.772.1————
Swin-LFramework=Mask2Former [5], Win. Size=12 x 12, # of Params=216 M, FLOPs=641 G, Pre-trained=ImageNet-22K, Resolution=800x8002022.0950.1—29.953.972.1————
DAT-S++FLOPs=430G, #Param=74M, Schedule=3x, Resolution=1280x800x32023.0945.769.927.649.264.349.7———
DAT-T++FLOPs=301G, #Param=45M, Schedule=3x, Resolution=1280x800x32023.0945.169.226.748.56448.7———
CSWin-BFLOPs=526G, #Param=97M, Schedule=3x, Resolution=1280x800x32023.0944.969.1———48.3———
InternImage-BFLOPs=501G, #Param=115M, Schedule=3x, Resolution=1280x800x32023.0944.868.726.847.663.148———
DAT-S++FLOPs=430G, #Param=74M, Schedule=1x, Resolution=1280x800x32023.0944.568.7254863.348.2———
InternImage-SFLOPs=340G, #Param=69M, Schedule=3x, Resolution=1280x800x32023.0944.568.524.847.762.947.8———
CSWin-SFLOPs=342G, #Param=54M, Schedule=3x, Resolution=1280x800x32023.0944.568.4———47.7———
InternImage-BFLOPs=501G, #Param=115M, Schedule=1x, Resolution=1280x800x32023.094467.824.347.262.547.4———
DAT-SFLOPs=378G, #Param=69M, Schedule=3x, Resolution=1280x800x32023.09446827.847.759.547.5———
CSWin-BFLOPs=526G, #Param=97M, Schedule=1x, Resolution=1280x800x32023.0943.967.8———47.3———
DiNAT-SFLOPs=330G, #Param=70M, Schedule=3x, Resolution=1280x800x32023.0943.96824.447.163.147.4———
S2DPrecision=FP162026.0243.8————————
DAT-T++FLOPs=301G, #Param=45M, Schedule=1x, Resolution=1280x800x32023.0943.767.924.547.462.447.3———
InternImage-TFLOPs=270G, #Param=49M, Schedule=3x, Resolution=1280x800x32023.0943.767.324.34762.547.3———
CSWin-TFLOPs=279G, #Param=42M, Schedule=3x, Resolution=1280x800x32023.0943.667.9———46.6———
AdamWPrecision=FP162026.0243.6————————
DiNAT-TFLOPs=258G, #Param=48M, Schedule=3x, Resolution=1280x800x32023.0943.567.224.246.962.346.7———
RegionViT-B+FLOPs=307G, #Param=93M, Schedule=3x, Resolution=1280x800x32023.0943.567.125.146.462.147———
ConvNeXt-BFLOPs=486G, #Param=108M, Schedule=3x, Resolution=1280x800x32023.0943.567.1———46.7———
RegionViT-S+FLOPs=183G, #Param=51M, Schedule=3x, Resolution=1280x800x32023.0943.467.124.346.46247———
InternImage-SFLOPs=340G, #Param=69M, Schedule=1x, Resolution=1280x800x32023.0943.367.123.146.862.546.7———
Swin-SFLOPs=359G, #Param=69M, Schedule=3x, Resolution=1280x800x32023.0943.367.328.146.758.646.6———
CSWin-SFLOPs=342G, #Param=54M, Schedule=1x, Resolution=1280x800x32023.0943.267.1———46.2———
NAT-SFLOPs=330G, #Param=70M, Schedule=3x, Resolution=1280x800x32023.0943.266.923.346.661.846.4———
S2DQuantization Method=ERQ, Bit-width (W/A)=W8A82026.0243.2————————
ConvNeXt-SFLOPs=348G, #Param=70M, Schedule=3x, Resolution=1280x800x32023.0942.966.9———46.2———
S2DQuantization Method=PTQ4ViT, Bit-width (W/A)=W8A82026.0242.8————————
ConvNeXt-BFLOPs=486G, #Param=108M, Schedule=1x, Resolution=1280x800x32023.0942.766.3———46———
PVTv2-B4FLOPs=500G, #Param=82M, Schedule=1x, Resolution=1280x800x32023.0942.766.123.345.66246.1———
NAT-TFLOPs=258G, #Param=48M, Schedule=3x, Resolution=1280x800x32023.0942.66623.145.761.445.9———
InternImage-TFLOPs=270G, #Param=49M, Schedule=1x, Resolution=1280x800x32023.0942.566.122.745.761.245.8———
PVTv2-B5FLOPs=557G, #Param=102M, Schedule=1x, Resolution=1280x800x32023.0942.565.722.445.861.146———
DAT-SFLOPs=378G, #Param=69M, Schedule=1x, Resolution=1280x800x32023.0942.566.725.545.858.545.4———
DAT-TFLOPs=272G, #Param=48M, Schedule=3x, Resolution=1280x800x32023.0942.466.127.245.857.145.5———
CSWin-TFLOPs=279G, #Param=42M, Schedule=1x, Resolution=1280x800x32023.0942.265.6———45.4———
AdamWQuantization Method=ERQ, Bit-width (W/A)=W8A82026.0241.9————————
ConvNeXt-SFLOPs=348G, #Param=70M, Schedule=1x, Resolution=1280x800x32023.0941.865.2———45.1———
ConvNeXt-TFLOPs=262G, #Param=48M, Schedule=3x, Resolution=1280x800x32023.0941.76521.944.859.845———
Swin-TFLOPs=267G, #Param=48M, Schedule=3x, Resolution=1280x800x32023.0941.665.125.945.156.944.9———
RegionViT-B+FLOPs=307G, #Param=93M, Schedule=1x, Resolution=1280x800x32023.0941.665.221.644.460.544.8———
S2DQuantization Method=ERQ, Bit-width (W/A)=W6A62026.0241.4————————
PVTv2-B2FLOPs=309G, #Param=45M, Schedule=1x, Resolution=1280x800x32023.0941.264.22243.759.444.4———
Swin-SFLOPs=359G, #Param=69M, Schedule=1x, Resolution=1280x800x32023.0941.164.925.144.356.644.2———
RegionViT-S+FLOPs=183G, #Param=51M, Schedule=1x, Resolution=1280x800x32023.0940.964.121.643.458.944———
ScaleKDFramework=Mask R-CNN, Backbone=Swin-T2024.1140.8—21.843.759.8————
DAT-TFLOPs=272G, #Param=48M, Schedule=1x, Resolution=1280x800x32023.0940.464.223.943.855.543.1———
ConvNeXt-TFLOPs=262G, #Param=48M, Schedule=1x, Resolution=1280x800x32023.0940.163.3———42.8———
AdamWQuantization Method=ERQ, Bit-width (W/A)=W6A62026.0239.9————————
Swin-TFLOPs=267G, #Param=48M, Schedule=1x, Resolution=1280x800x32023.0939.863.324.243.154.642.7———
BaselineFramework=Mask R-CNN, Backbone=Swin-T2024.1139.3—20.541.857.8————
ScaleKDFramework=Mask R-CNN, Backbone=ResNet-502024.1139.1—19.342.557.1————
BaselineFramework=Mask R-CNN, Backbone=ResNet-502024.1137.1—1840.154.9————
AdamWQuantization Method=PTQ4ViT, Bit-width (W/A)=W8A82026.0236.5————————
SAM FTI-FDet2026.033656.6——————49.5
RSPrompter2026.0335.254.3——————51
S2DQuantization Method=ERQ, Bit-width (W/A)=W5A52026.0234.4————————
SAM-seg2026.0333.253.3——————42
FastSAM2026.0332.651.8——————49
S2DQuantization Method=PTQ4ViT, Bit-width (W/A)=W6A62026.0220.7————————
S2DQuantization Method=ERQ, Bit-width (W/A)=W4A42026.0213————————
AdamWQuantization Method=ERQ, Bit-width (W/A)=W5A52026.0211.7————————
AdamWQuantization Method=PTQ4ViT, Bit-width (W/A)=W6A62026.021.9————————
AdamWQuantization Method=ERQ, Bit-width (W/A)=W4A42026.020.5————————
ConvNeXt-TBackbone=ConvNeXt-T, #param.=48M, FLOPs=262G, Framework=Mask R-CNN 1x2024.07———40.1——63.342.8—
GroupMamba-TBackbone=GroupMamba-T, #param.=40M, FLOPs=279G, Framework=Mask R-CNN 1x2024.07———42.9——66.546.3—
LocalVMamba-TBackbone=LocalVMamba-T, #param.=45M, FLOPs=291G, Framework=Mask R-CNN 1x2024.07———42.2——65.745.5—
ResNet-50Backbone=ResNet-50, #param.=44M, FLOPs=260G, Framework=Mask R-CNN 1x2024.07———34.7——55.737.2—
Swin-TBackbone=Swin-T, #param.=48M, FLOPs=267G, Framework=Mask R-CNN 1x2024.07———39.3——62.242.2—
VMamba-TBackbone=VMamba-T, #param.=50M, FLOPs=270G, Framework=Mask R-CNN 1x2024.07———42.7——66.346—