List your company
ResearchFormerly WizWand

Semantic Segmentation on Cityscapes (val) (mIoU)

87.1mIoUHMS + DejaVu
78.98881.09483.285.306Nov 28, 2018Feb 24, 2020May 23, 2021Aug 20, 2022Nov 16, 2023Feb 12, 2025May 12, 2026
Updated 4mo ago

Evaluation results

MethodLinks
HMS + DejaVuBackbone=HRNet48, GMacs=8932023.0387.1
OursHead=Upernet [42], Training data=Mapi+City, Params=912M2023.1287.1
InternImage-HBackbone=InternImage-H2022.1187
InternImage-HHead=Mask2Former [5], Training data=Mapi+City, Params=1.2B2023.1287
PSA2022.1186.9
HRNet-OCR(PSA)Head=OCRHead [48], Training data=Mapi+City2023.1286.9
HMSBackbone=HRNet48, GMacs=8932023.0386.7
OCR + DejaVuBackbone=HRNet48, GMacs=3482023.0386.5
InternImage-XLHead=Upernet [42], Training data=Mapi+City, Params=368M2023.1286.4
HRNet-OCR(MA)Head=OCRHead [48], Training data=Mapi+City2023.1286.3
OCRBackbone=HRNet48, GMacs=3482023.0386.1
OneFormerHead=Transformer decoder [15], Training data=Mapi+City, Params=372M2023.1285.8
ViT-Adapte-LHead=Mask2Former [5], Training data=Mapi+City, Params=571M2023.1285.8
OursHead=Upernet [42], Training data=City, Params=912M2023.1285.7
HRNet + DejaVuBackbone=HRNet48, GMacs=1752023.0385.4
SeMaskHead=Mask2Former [5], Training data=City, Params=222M2023.1285
ViT AdapterBackbone=ViT, GMacs=10892023.0384.9
HRNetBackbone=HRNet48, GMacs=1752023.0384.7
OneFormerHead=Transformer decoder [15], Training data=City, Params=372M2023.1284.6
OpenSeeD (L)Type=Open-vocabulary, Backbone=Large2023.0384.5
Mask2FormerHead=Mask2Former [5], Training data=City2023.1284.3
SegformerBackbone=MiT-B5, GMacs=3622023.0384
SeMaskBackbone=Swin-L, GMacs=2582023.0384
SegFormerHead=SegFormer [43], Training data=Mapi+City, Params=85M2023.1284
DDPHead=DeformableHead [17], Training data=City, Params=209M2023.1283.9
Image Resynthesis++Retrain=false2021.0383.5
OursRetrain=false2021.0383.5
Mask2FormerBackbone=Swin-L, GMacs=2512023.0383.3
HRFormer-B + OCR + SegFix#params.=56.2M, FLOPs=1119.9G, Multi-scale testing=true, Extra pre-training on ADE20K=false2021.1083.2
OneFormer (L)Type=Closed-set, Backbone=Large2023.0383
Mask2Former (L)Type=Closed-set, Backbone=Large2023.0382.9
DDPSHead=SegFormer [43], Training data=City, Params=123M2023.1282.9
DeeplabV3Backbone=ResNeSt-2002020.0482.7
Mask2Former (B)Type=Closed-set, Backbone=Base2023.0382.7
WaveMixPre-train=ImageNet-1k, Parameters=63 M, Evaluation=Single-scale2022.0582.7
HRFormer-B + OCR#params.=56.2M, FLOPs=1119.9G, Multi-scale testing=true, Extra pre-training on ADE20K=false2021.1082.6
SegFormer-B5Encoder Size=81.4M2022.0482.4
Previous SOTA [1]Pre-train=ImageNet-1k, Parameters=85 M, Evaluation=Single-scale2022.0582.4
FAN-L-HybridEncoder Size=76.8M2022.0482.3
SETR-PUP#params.=317.8M, FLOPs=2326.7G, Multi-scale testing=true, Extra pre-training on ADE20K=false2021.1082.2
FAN-B-HybridEncoder Size=50.4M2022.0482.2
SETRHead=SETRUPHead [51], Training data=City, Params=318M2023.1282.2
SETR-PUPBackbone=T-Large, Training schedule=80k, Inference scale=Multi-scale (MS)2020.1282.15
CleanBackbone=Swin-T2024.1082.1
DANetBackbone=ResNet-1012021.0582
OCR + DejaVuBackbone=HRNet18, GMacs=392023.0382
SA-Gate + BMPDepth Data=true, Backbone=ResNet-1012020.0781.7
EANetBackbone=ResNet-1012021.0581.7
X-Decoder (L)Type=Open-vocabulary, Backbone=Large2023.0381.7
SETR-PUPBackbone=T-Large, Training schedule=40k, Inference scale=Multi-scale (MS)2020.1281.57
DANetDepth Data=false, Backbone=ResNet-1012020.0781.5
ACFNetDepth Data=false, Backbone=ResNet-1012020.0781.5
Axial-DeepLab-LBackbone=Axial-ResNet-L, Inference scale=Multi-scale (MS)2020.1281.5
FAN-S-HybridEncoder Size=26.3M2022.0481.5
HS3-FuseBackbone=HRNet18, GMacs=392023.0381.4
CCNetDepth Data=false, Backbone=ResNet-1012020.0781.3
FAN-T-HybridEncoder Size=7.4M2022.0481.2
HRNetV2-W48Backbone=HRNet, Model Size=65.9M2020.0681.1
Axial-DeepLab-XLBackbone=Axial-ResNet-XL, Inference scale=Multi-scale (MS)2020.1281.1
EMANetBackbone=ResNet-1012021.0581
PSPNetBackbone=ResNet-1012021.0581
SegFormer-B2Encoder Size=24.2M2022.0481
PSANetBackbone=ResNet-1012021.0580.9
DDFPLabeled image ratio=1/2 (1488)2024.0380.82
DPCBackbone=Xception-71, Extra COCO dataset for training=true2018.1180.8
GSCNNDepth Data=false, Backbone=WideResNet-1012020.0780.8
SA-Gate + BMPDepth Data=true, Backbone=ResNet-502020.0780.7
DMNetBackbone=ResNet-1012021.0580.7
GCNetBackbone=ResNet-1012021.0580.7
Deeplabv3#params.=87.1M, FLOPs=1394.0G, Multi-scale testing=true, Extra pre-training on ADE20K=false2021.1080.7
OCRBackbone=HRNet18, GMacs=392023.0380.7
CCNetBackbone=ResNet-1012018.1180.5
RGB baselineDepth Data=false, Backbone=ResNet-1012020.0780.5
RGB-D baselineDepth Data=true, Backbone=ResNet-1012020.0780.5
Mask2Former (T)Type=Closed-set, Backbone=Tiny2023.0380.5
Mask2FormerBackbone=Swin-T, Param=47.4M2026.0580.5
AugSegLabeled image ratio=1/2 (1488)2024.0380.43
DeeplabV3Backbone=ResNeSt-1012020.0480.42
MDEQ-XLBackbone=MDEQ, Model Size=70.9M2020.0680.3
Softmax EntropyRetrain=false2021.0380.3
Embedding DensityRetrain=false2021.0380.3
ANNBackbone=ResNet-1012021.0580.3
OpenSeeD (T)Type=Open-vocabulary, Backbone=Tiny2023.0380.3
CPSLabeled image ratio=1/2 (1488)2024.0380.21
CCNetBackbone=ResNet-1012020.1280.2
MambaPanopticBackbone=SegMan, Param=35.7M2026.0580.18
OCNetBackbone=ResNet1012020.0480.1
BiAlignNetgamma (down-sampling ratio)=1, Backbone=DFNet2, FPS=32, Parameters=19.2M, Pre-training=Mapillary dataset, Hardware=1080Ti GPU2021.0580.1
PSPNet#params.=68.0M, FLOPs=1028.8G, Multi-scale testing=true, Extra pre-training on ADE20K=false2021.1080
APCNetBackbone=ResNet-1012021.0579.9
DDFPLabeled image ratio=1/4 (744)2024.0379.88
DeeplabV3Backbone=ResNeSt-502020.0479.87
AugSegLabeled image ratio=1/4 (744)2024.0379.56
UniMatchLabeled image ratio=1/2 (1488)2024.0379.5
DeeplabV3Backbone=ResNet1012020.0479.42
CleanBackbone=ResNet502024.1079.4
SETR-PUPBackbone=T-Large, Training schedule=80k, Inference scale=Single-scale (SS)2020.1279.34
DeepLabv3Backbone=ResNet-1012018.1179.3
DeeplabV3Backbone=ResNet1012020.0479.3
DeepLab-v3Backbone=ResNet-101, Inference scale=Multi-scale (MS)2020.1279.3