List your company
ResearchFormerly WizWand

Semantic Segmentation on COCO (mIoU)

67.4mIoUFull-precision
22.57634.21345.8557.487Oct 18, 2022May 28, 2023Jan 6, 2024Aug 16, 2024Mar 26, 2025Nov 4, 2025Jun 15, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Full-precisionBits (W/A)=16/162026.0667.4
Shift-and-Sum Quantization + LiteVARBits (W/A)=6/62026.0667
LiteVARBits (W/A)=6/62026.0666.9
HIPIEBackbone=ViT-H2023.0766.8
HIPIEBackbone=ViT-H2023.0766.8
Shift-and-Sum QuantizationBits (W/A)=6/62026.0666.8
BRECQBits (W/A)=6/62026.0666.5
Labeled OnlyEncoder=DINOV2-B, #Params=97.5M2024.1066.4
X-DecoderBackbone=DaViT-B2023.0766
X-DecoderBackbone=DaViT-B2023.0766
SEEMBackbone=DaViT-B2023.0765.3
SEEMBackbone=DaViT-B2023.0765.3
ODISEBackbone=ViT-H+SD2023.0765.2
ODISEBackbone=ViT-H+SD2023.0765.2
Shift-and-Sum Quantization + LiteVARBits (W/A)=4/42026.0665
Shift-and-Sum QuantizationBits (W/A)=4/42026.0664.6
LiteVARBits (W/A)=4/42026.0664.4
BRECQBits (W/A)=4/42026.0663.7
DaTaSegBackbone=ViTDet-L, Training data=COCO panoptic + ADE semantic + O365 bbox2023.0662.9
DiveUpV.A. (VFM-agnostic)=true, Backbone=DINOv2-S, Evaluation Protocol=Linear Probing, Refinement stage=false2026.0362.71
DaTaSegBackbone=ViTDet-B, Training data=COCO panoptic + ADE semantic2023.0662.7
Labeled OnlyEncoder=DINOV2-S, #Params=24.8M2024.1062.5
X-DecoderBackbone=FocalT2023.0762.4
X-DecoderBackbone=FocalT2023.0762.4
LoftUpV.A. (VFM-agnostic)=false, Backbone=DINOv2-S, Evaluation Protocol=Linear Probing, Refinement stage=false2026.0362.23
NAFV.A. (VFM-agnostic)=true, Backbone=DINOv2-S, Evaluation Protocol=Linear Probing, Refinement stage=true2026.0362.18
SEEMBackbone=FocalT2023.0761.2
SEEMBackbone=FocalT2023.0761.2
AnyUp-SV.A. (VFM-agnostic)=true, Backbone=DINOv2-S, Evaluation Protocol=Linear Probing, Refinement stage=false2026.0361.1
JAFARV.A. (VFM-agnostic)=false, Backbone=DINOv2-S, Evaluation Protocol=Linear Probing, Refinement stage=false2026.0360.78
FeatUpV.A. (VFM-agnostic)=false, Backbone=DINOv2-S, Evaluation Protocol=Linear Probing, Refinement stage=false2026.0360.1
HIPIEBackbone=RN502023.0759.5
HIPIEBackbone=RN502023.0759.5
BilinearV.A. (VFM-agnostic)=true, Backbone=DINOv2-S, Evaluation Protocol=Linear Probing, Refinement stage=false2026.0359.03
DaTaSegBackbone=R-50, Training data=COCO panoptic2023.0657.7
ODISEBackbone=UNet+M2F, Training data=LAION+CLIP+COCO2023.0652.4
UniMatch + ECOCSegLabel Ratio=1/322025.1251.6
AllSparkLabel Ratio=1/322025.1250.9
DiGSeginput size=512²2026.0450.8
UniMatch + ECOCSegLabel Ratio=1/642025.1249.9
UniMatchLabel Ratio=1/322025.1249.8
AllSparkLabel Ratio=1/642025.1249.5
EoMTinput size=512²2026.0448.7
MSegBackbone=HRNet-W48, Venue=CVPR 20, Label space=MR2024.0748.6
UniMatchLabel Ratio=1/642025.1248.2
SegMANinput size=512²2026.0448.2
4M-21Model Scale=XL, Evaluation Protocol=Out-of-the-box2024.0648.1
VWFormer-B5input size=512²2026.0448
AutoUniSeg (Ours)Backbone=HRNet-W48, Label space=Auto2024.0746.7
SegFormer-B5input size=512²2026.0446.7
4M-21Model Scale=Large, Evaluation Protocol=Out-of-the-box2024.0646.4
UniMatch + ECOCSegLabel Ratio=1/1282025.1246.2
PC2SegLabel Ratio=1/322025.1246.1
CGRSeg-Linput size=512²2026.0446
OffSeg-Linput size=512²2026.0446
AllSparkLabel Ratio=1/1282025.1245.4
UniMatchLabel Ratio=1/1282025.1244.4
Unified-IOModel Scale=XL, Evaluation Protocol=Out-of-the-box2024.0644.3
PC2SegLabel Ratio=1/642025.1243.7
PseudoSegLabel Ratio=1/322025.1243.6
4M-21Model Scale=Base, Evaluation Protocol=Out-of-the-box2024.0642.5
Sup-onlyLabel Ratio=1/322025.1242.2
PseudoSegLabel Ratio=1/642025.1241.8
UniMatch + ECOCSegLabel Ratio=1/2562025.1241.8
Unified-IO 2Model Scale=XXL, Evaluation Protocol=Out-of-the-box2024.0641.7
AllSparkLabel Ratio=1/2562025.1241.6
Unified-IOModel Scale=Large, Evaluation Protocol=Out-of-the-box2024.0641.6
PC2SegLabel Ratio=1/1282025.1240.1
Unified-IO 2Model Scale=XL, Evaluation Protocol=Out-of-the-box2024.0639.7
Uni NLL+Backbone=SNp-DN161, Venue=IJCV 24, Label space=MC2024.0739.3
PseudoSegLabel Ratio=1/1282025.1239.1
UniMatchLabel Ratio=1/2562025.1238.9
Unified-IO 2Model Scale=Large, Evaluation Protocol=Out-of-the-box2024.0638.9
OpenSegBackbone=Eff-b7, Training data=COCO+Loc. Narr.2023.0638
Single datasetBackbone=HRNet-W48, Label space=DS2024.0738
Sup-onlyLabel Ratio=1/642025.1237.8
PC2SegLabel Ratio=1/2562025.1237.5
PseudoSegLabel Ratio=1/2562025.1237.1
Multi-SegHeadBackbone=HRNet-W48, Label space=DS2024.0736.7
OpenSegBackbone=R-101, Training data=COCO pan+cap2023.0636.1
Auto univ.Backbone=SNp-RN18, Venue=BMVC 22, Label space=Auto2024.0735.6
NLL+Backbone=SNp-RN18, Venue=WACV 22, Label space=MC2024.0735.4
CLS-SEGMethod Category=CLIP-based, Backbone=ViT-B/16, Post-processing=denseCRF2023.1235.3
UniMatch + ECOCSegLabel Ratio=1/5122025.1234.5
AllSparkLabel Ratio=1/5122025.1234.1
CLS-SEGMethod Category=CLIP-based, Backbone=ViT-B/162023.1234
Sup-onlyLabel Ratio=1/1282025.1233.6
Unified-IOModel Scale=Base, Evaluation Protocol=Out-of-the-box2024.0632.9
UniMatchLabel Ratio=1/5122025.1231.9
PC2SegLabel Ratio=1/5122025.1229.9
PseudoSegLabel Ratio=1/5122025.1229.8
OpenSDTraining Data=ADE, Backbone=Swin-T2023.1229.1
Sup-onlyLabel Ratio=1/2562025.1228
NamedMaskMethod Category=CLIP-based2023.1227.7
SegCLIPArch.=ViT, Init.=true, Training Data=CC+COCO, Sup.=Text, Zero-Shot=true2022.1126.5
SegCLIPMethod Category=CLIP-based2023.1226.5
CLIPpyArch=ViT, Dataset=HQITP-134M, SSP=true2022.1025.5
CLIPSurgeryMethod Category=CLIP-based, Implementation=Re-implemented2023.1225.2
GroupViTArch.=ViT, Init.=false, Training Data=CC12M+YFCC, Sup.=Text, Zero-Shot=true2022.1124.3
GroupViTMethod Category=CLIP-based2023.1224.3