List your company
ResearchFormerly WizWand

Semantic Segmentation on Cityscapes (val) (mIoU, mIoU (+MS))

0.833mIoUMask2Former
0.789320.800660.8120.82334Mar 30, 2023
Updated 8mo ago

Evaluation results

MethodLinks
Mask2FormerBackbone=Swin-L, Pre-training=ImageNet-22K, #Param=216M, FLOPs=2113G2023.030.8330.843
DDPBackbone=ConvNext-L, Sampling steps=3, Pre-training=ImageNet-22K, #Param=209M, FLOPs=3245G2023.030.83210.8392
DDPBackbone=ConvNext-L, Sampling steps=1, Pre-training=ImageNet-22K, #Param=209M, FLOPs=2139G2023.030.82950.8376
DDPBackbone=ConvNext-B, Sampling steps=3, #Param=100M, FLOPs=2447G2023.030.82780.8349
DiversePatchBackbone=Swin-L, Pre-training=ImageNet-22K, #Param=234M, FLOPs=3190G2023.030.8270.836
DDPBackbone=ConvNext-S, Sampling steps=3, #Param=62M, FLOPs=2166G2023.030.82690.8358
DDPBackbone=ConvNext-T, Sampling steps=3, #Param=40M, FLOPs=1989G2023.030.8260.8315
DDPBackbone=ConvNext-B, Sampling steps=1, #Param=100M, FLOPs=1340G2023.030.82590.8347
DDPBackbone=Swin-B, Sampling steps=3, #Param=99M, FLOPs=2464G2023.030.82540.8342
DDPBackbone=Swin-S, Sampling steps=3, #Param=61M, FLOPs=2174G2023.030.82410.8321
DDPBackbone=Swin-B, Sampling steps=1, #Param=99M, FLOPs=1357G2023.030.82370.8336
DDPBackbone=ConvNext-S, Sampling steps=1, #Param=62M, FLOPs=1059G2023.030.82370.8338
DDPBackbone=ConvNext-T, Sampling steps=1, #Param=40M, FLOPs=883G2023.030.82330.83
SegFormer-B5Backbone=MiT-B5, #Param=85M, FLOPs=1448G2023.030.82250.8348
DDPBackbone=Swin-S, Sampling steps=1, #Param=61M, FLOPs=1067G2023.030.82170.8306
OCRNetBackbone=HRFormer-B, #Param=56M, FLOPs=2240G2023.030.8190.826
DDPBackbone=Swin-T, Sampling steps=3, #Param=39M, FLOPs=1992G2023.030.81240.8246
DDPBackbone=Swin-T, Sampling steps=1, #Param=39M, FLOPs=885G2023.030.80960.8225
StructTokenBackbone=ViT-L, Pre-training=ImageNet-22K, #Param=364M, FLOPs=2913G2023.030.80050.8207
SETR-PUPBackbone=ViT-L, Pre-training=ImageNet-22K, #Param=318M, FLOPs=2955G2023.030.79340.8215
SegmenterBackbone=ViT-L, #Param=333M, FLOPs=2685G2023.030.7910.813